Received: 2 May 2021 Revised: 10 July 2021 Accepted: 1 September 2021

DOI: 10.1002/cac2.12215

REVIEW

Artificial intelligence for assisting cancer diagnosis and

treatment in the era of precision medicine
Zi-Hang Chen1,2,† Li Lin1,† Chen-Fei Wu1 Chao-Feng Li3 Rui-Hua Xu4
Ying Sun1
1Department of Radiation Oncology, State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine,
Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong
510060, P. R. China
2 Zhongshan School of Medicine, Sun Yat-sen University, Guangzhou, Guangdong 510080, P. R. China
3Artificial Intelligence Laboratory, State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine,
Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong
510060, P. R. China
4 Department of Medical Oncology, State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine,

Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong
510060, P. R. China

Correspondence
Ying Sun, Department of Radiation Oncol- Abstract
ogy, Sun Yat-sen University Cancer Center, Over the past decade, artificial intelligence (AI) has contributed substantially
State Key Laboratory of Oncology in South
China, Collaborative Innovation Center to the resolution of various medical problems, including cancer. Deep learning
of Cancer Medicine, #651 Dongfeng Road (DL), a subfield of AI, is characterized by its ability to perform automated fea-
East, Guangzhou 510060, Guangdong, P. R.
ture extraction and has great power in the assimilation and evaluation of large
China.
Email: [email protected] amounts of complicated data. On the basis of a large quantity of medical data
Rui-Hua Xu, Department of Medical and novel computational technologies, AI, especially DL, has been applied in
Oncology, Sun Yat-sen University Cancer
various aspects of oncology research and has the potential to enhance cancer
Center, State Key Laboratory of Oncology
in South China, Collaborative Innovation diagnosis and treatment. These applications range from early cancer detection,
Center for Cancer Medicine, Guangdong diagnosis, classification and grading, molecular characterization of tumors, pre-
Key Laboratory of Nasopharyngeal Carci-
noma Diagnosis and Therapy, Guangzhou
diction of patient outcomes and treatment responses, personalized treatment,
510060, Guangdong, P. R. China. automatic radiotherapy workflows, novel anti-cancer drug discovery, and clini-
Email: [email protected] cal trials. In this review, we introduced the general principle of AI, summarized
major areas of its application for cancer diagnosis and treatment, and discussed

Abbreviations: AI, artificial intelligence; ML, machine learning; DL, deep learning; DNN, deep neural network; CNN, convolutional neural network;
CIN, intra-epithelial neoplasia; DS, dual-stained; AUC, area under the curve; ctDNA, circulating tumor DNA; cfDNA, cell-free DNA; HE,
hematoxylin-eosin; PET-CT, positron emission tomography-CT; WSI, whole slide imaging; CT, computed tomography; MRI, magnetic resonance
imaging; 3D, three dimensional; NPC, nasopharyngeal carcinoma; GRAIDS, gastrointestinal AI diagnostic system; MSI, microsatellite instability;
TMB, tumor mutational burden; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; pCR, complete pathologic response; OAR,
organs at risk; GTV, gross tumor volume; CTV, clinical target volume; RCT, randomized controlled trial; Fbw7, F-box/WD repeat-containing protein 7;
DDR1, discoidin domain receptor 1; DDL, distributed deep learning; FDA, Food and Drug Administration; NAC, neoadjuvant chemotherapy; DSC,
Dice similarity coefficient

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium,
provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2021 The Authors. Cancer Communications published by John Wiley & Sons Australia, Ltd. on behalf of Sun Yat-sen University Cancer Center

1100 wileyonlinelibrary.com/journal/cac2 Cancer Communications. 2021;41:1100–1115.

 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1101

† These authors contributed equally to this

work. its future directions and remaining challenges. As the adoption of AI in clinical
use is increasing, we anticipate the arrival of AI-powered cancer care.

KEYWORDS
artificial intelligence, cancer diagnosis, cancer research, cancer treatment, convolutional neu-
ral network, deep learning, deep neural network, oncology

1 BACKGROUND research within the oncology field involves the utilization

of DL.
At a workshop in Dartmouth in the summer of 1956, Among DNN models, convolutional neural networks
McCarthy et al. [1] coined the term “artificial intelligence (CNNs) are the most popular DL architectures. They have
(AI)”, also known as “machine intelligence”. To put it sim- been used for cancer lesion detection, recognition, seg-
ply, AI is defined as a programmed machine that can mentation and the classification of medical images [8–10].
learn and recognize patterns and relationships between The architecture of a typical CNN (Figure 1) is structured
inputs and outputs and use this knowledge effectively for by stacking three main layers: convolutional layers, pool-
decision-making on brand-new input data [1, 2]. Machine ing layers, and fully-connected layers. In doing this, CNNs
learning (ML) and deep learning (DL) are the predom- transform the original images layer by layer from pixel val-
inant methods used to actualize AI and are sometimes ues to the final prediction scores. The convolutional lay-
used synonymously. In the field of computer science, ML ers involve combining input data (feature map) with con-
is a subfield of AI, and DL is a specific subset of ML volutional kernels (filters) to form a transformed feature
that focuses on deep artificial neural networks (Figure 1). map. The filters in the convolutional layers are automat-
Over the past decade, following advances in big data, algo- ically adjusted based on learned parameters to extract the
rithms, computer power, and internet technology, DL has most useful features for a specific task. Yet, there is a draw-
achieved unprecedented success in various tasks in vari- back; it is difficult to tell what features are learned by the
ous fields, including facial recognition, image classifica- CNNs, which is known as the “black box”.
tion, voice recognition, automatic translation, and health- Over the past five years, large amounts of researches
care [3]. Given the great number of patients diagnosed with have applied DL to cancer diagnosis, precision medicine,
cancers each year worldwide [4], there is an acute inter- radiotherapy, and cancer research (Figure 2). Moreover,
est in the application of AI in oncology, and such interests the American Food and Drug Administration (FDA) have
include making accurate diagnosis of cancers using patho- approved a number of AI algorithms related to oncology
logical slides and radiological images, predicting patient (Table 1) and published a fast-track approval plan for AI
outcomes, and optimizing treatment decisions. AI there- medical algorithms in 2018. Here, we provided an overview
fore has the potential to solve the problem of unbalanced of the recent and enormous progresses in the application of
distribution of medical resources and improve cancer AI in oncology in this review (Figure 3). We also highlight
care. the limitations, challenges, and future implications of AI-
Inspired by brain neural architecture, DL uses deep neu- powered cancer care.
ral networks (DNNs) to develop sophisticated models with
multiple hidden layers to analyze various types of data
and develop prediction outputs (Figure 1) [5]. Unlike con- 2 CANCER SCREENING, DIAGNOSIS,
ventional ML techniques, which require careful engineer- CLASSIFICATION, AND GRADING
ing to design a feature extractor that transforms raw data
(such as the pixel values of an image) into relevant dis- Cancer screening for early detection, accurate cancer diag-
criminatory features before data input, DL algorithms feed nosis, classification and grading are the key determinants
the machine with raw data with which it can automat- of treatment decisions and patient outcomes. Over the past
ically learn the optimal deep features that best fit the few years, there is increasing interest in the applications
task through a training process [6, 7]. This ability likely of AI in these critical areas (Table 2), sometimes with per-
explains the fact that DL algorithms have been consistently formance equivalent to human experts and advantages
improved in many common AI tasks, such as image recog- in scalability and time-saving. More importantly, AI has
nition, pattern recognition, speech recognition, and nat- shown its potential in solving challenging problems that
ural language processing. Consequently, a majority of AI humans simply cannot do.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1102 CHEN et al.

F I G U R E 1 The relationship between artificial intelligence, machine learning, and deep learning and commonly used algorithms as
examples. CNN, convolutional neural network

2.1 Cancer screening and early orectal cancer screening) where the treatment leads to a
detection decrease in the incidence of invasive cancer [13]. Given
the requirement for high throughput technology and a fast
Cancer screening has contributed to decreasing the mor- turnaround, automation is being used to improve the effi-
tality of some common cancers [11, 12]. The most suc- ciency of cancer screening.
cessful examples are the identification of precancerous For cervical cancer screening, Wentzensen et al. [14]
lesions (e.g., cervical intra-epithelial neoplasia [CIN] for developed a DL classifier for p16/Ki-67 dual-stained (DS)
cervical cancer screening, and adenomatous polyps for col- cytology slides trained on biopsy-based gold standards.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1103

< 0.001), while it had a similar performance in identi-

fying high-grade CIN, which indicates immediate treat-
ment. For colorectal cancer screening, a prospective ran-
domized controlled trial including 1,058 patients showed
that AI-assisted colonoscopy significantly increased ade-
noma detection rates and the mean number of ade-
nomas found per patient compared with conventional
colonoscopy (29.1% vs. 20.3%), which was attributed to a
higher number of diminutive adenomas found [15]. This
is particularly important because a 1% increase in the ade-
noma detection rate is associated with a 3% decrease in col-
orectal cancer incidence [13].
Automated nodule detection and classification on low-
dose computed tomography (CT) and mammography for
lung and breast cancer screening have attracted signifi-
cant attention. Several successful CNN-based models have
achieved classification accuracies of 80% to 95% [16–18],
which shows their transformative potential in lung cancer
screening. Ardila et al. [19] proposed a DL algorithm that
uses patients current and prior low-dose CT scans to pre-
dict the risk of lung cancer with outstanding results (area
under the curve [AUC] of receiver operating characteristic
= 0.944). Improvement in breast cancer screening with AI
F I G U R E 2 Publication statistics of deep learning by cancer mammography has also been verified in preclinical studies
area over the past five years, searched on PubMed. A. Publication [20–24], as well as in clinical settings [25]. McKinney et al.
statistics of deep learning by cancer diagnosis, precision medicine, [25] established an AI system for breast cancer screening
radiotherapy, and cancer research. B. Publication statistics of deep
using an ensemble of three CNN-based models. A reduc-
learning for different cancer sites
tion in the numbers of false positives and false negatives
was observed compared with the original decisions made
In independent testing, AI-based DS had equal sensitiv- in the course of clinical practice. In an independent study
ity and substantially higher specificity compared with a by six radiologists, the AUC for the AI system was 11.5%
Pap smear and manual interpretation of DS. Most impor- higher than the average AUC achieved by the 6 radiolo-
tantly, AI-based DS reduced unnecessary colposcopies by gists. Notably, this AI system has the ability to generalize
one-third compared with Pap smears (41.9% vs. 60.1%, P from the training data to multicenter data.

TA B L E 1 Summary of FDA-approved artificial intelligence devices in the field of oncology

AI algorithm Company FDA approval date Indication
ClearRead CT Riverain Technologies 09/09/2016 Detection of pulmonary nodules
QuantX Quantitative Insights 07/19/2017 Diagnosing breast cancer
Arterys Oncology DL Arterys 01/25/2018 Liver and lung cancer diagnosis
cmTriage CureMetrix 03/08/2019 Detection of suspicious breast lesions
Koios DS Breast Koios Medical 07/03/2019 Breast lesion malignancy evaluation
ProFound AI Software V2.1 iCAD 10/04/2019 Breast lesion malignancy evaluation
Transpara ScreenPoint Medical BV 03/05/2020 Breast lesion malignancy evaluation
syngo.CT Lung CAD Siemens Healthcare GmbH 03/09/2020 Detection of pulmonary nodules
MammoScreen Therapixel 03/25/2020 Breast lesion malignancy evaluation
Rapid ASPECTS iSchema View 06/26/2020 Detection of suspicious brain lesions
InferRead Lung CT.AI InferRead Lung CT.AI 07/02/2020 Detection of pulmonary nodules
HealthMammo Zebra Medical Vision 07/16/2020 Detection of suspicious breast lesions
Abbreviations: FDA, Food and Drug Administration; AI, artificial intelligence; CT, computed tomography; DL, deep learning; CAD, computer-aided diagnosis.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1104 CHEN et al.

FIGURE 3 Applications of AI in cancer diagnosis, treatment and research. OARs, organs at risk

TA B L E 2 Summary of key papers applying deep learning to cancer diagnosis and treatment

Application Reference Task Performance

Screening
Pathology [14] Automation of dual stain cytology in cervical Sensitivity, 87%
cancer screening
Endoscopy [15] Automation of polyp detection False positive rate, 7.5%
Radiology [16] Predicting invasiveness of pulmonary AUC, 0.788
adenocarcinomas
Radiology [17] Lung nodule classification: benign/malignant Sensitivity, 98.45%
Radiology [18] Lung nodule classification: benign/malignant Accuracy, 79.5%
Radiology [19] Lung nodule classification: benign/malignant AUC, 0.944
Radiology [20] Breast lesion classification: benign/malignant AUC, 0.909
Radiology [21] Breast lesion classification: benign/malignant AUC, 0.860
Radiology [22] Breast lesion classification: benign/malignant AUC, 0.870
Radiology [23] Breast lesion classification: benign/malignant AUC, 0.860
Radiology [24] Breast lesion classification: benign/malignant AUC, 0.890
Radiology [25] Breast cancer prediction AUC, 0.8107
Diagnosis
Pathology [30] Invasive breast cancer detection DSC, 75.86%
Pathology [31] Breast cancer nodal metastasis detection AUC, 0.994
Pathology [32] Breast lesion classification: benign/malignant Accuracy, 98.7%
Pathology [33] Detection of lymph node metastases in breast AUC, 0.994
cancer
Pathology [35] Diagnosis of gastric cancer AUC, 0.990-0.996
Pathology [36] Predicting origins for cancers of unknown Accuracy, 80%
primary
(Continues)
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1105

TA B L E 2 (Continued)

Application Reference Task Performance

Pathology [51] Lung tumor classification: normal/ AUC, 0.97
adenocarcinoma/squamous cell carcinoma
Pathology [52] Automated Gleason grading of prostate Cohen’s quadratic kappa
adenocarcinoma statistic, 0.75
Radiology [37] Brain tumor classification: AUC, 0.984
normal/glioblastoma/sarcoma/metastatic
bronchogenic carcinoma
Radiology [38] Liver cancer detection Accuracy, 99.38%
Radiology [39] Prostate lesion classification: benign/malignant AUC, 0.84
Radiology [40] Detection of synchronous peritoneal Accuracy, 94.11%
carcinomatosis in colorectal cancer
Radiology [41] Detection of NPC using MRI Accuracy, 97.77%
Radiology [53] Predicting grade of liver cancer AUC, 0.83
Endoscopy [42] Gastric lesion classification: normal/malignant Accuracy, 96.49%
Endoscopy [43] Upper gastrointestinal cancer detection Accuracy, 99.7%
Endoscopy [44] Polyps identification Accuracy, 96%
Endoscopy [50] Polyps identification AUC, 0.984
Endoscopy [45] Invasive colorectal cancer diagnosis Accuracy, 94.1%
Endoscopy [46] Diminutive colorectal polyps classification: Accuracy, 90.1%
hyperplastic/neoplastic
Endoscopy [47] cT1b colorectal cancer diagnosis AUC, 0.871
Endoscopy [49] Nasopharyngeal lesion classification: Accuracy, 88%
benign/malignant
Prediction of mutation
Pathology [51] Predicting genetic mutations of lung cancer: AUC, 0.733-0.856
STK11, EGFR, FAT1, SETBP1, KRAS, and TP53
Pathology [56] Predicting genetic mutations of lung cancer: AUC>0.71
CTNNB1, FMN2, TP53, and ZFX4
Pathology [59] Predicting MSI status in colorectal cancer AUC, 0.93
Pathology [60] Predicting MSI status in colorectal cancer AUC, 0.85
Pathology [61] Predicting TMB status in gastric cancer AUC, 0.75
Pathology [61] Predicting TMB status in colon cancer AUC, 0.82
Radiology [62] Predicting EGFR status in NSCLC AUC, 0.81
Radiology [63] Predicting EGFR status in NSCLC AUC, 0.81
Radiology [70] Predicting TMB status in NSCLC AUC, 0.81
Predicting of prognosis
Pathology [66] Predicting outcome of colorectal cancer AUC, 0.69
Pathology [67] Predicting outcome of mesothelioma Concordance index, 0.643
Pathology [68] Predicting outcome of NSCLC AUC, 0.85
Immunotherapy
Radiology [70] Predicting response to immunotherapy in AUC, 0.81
advanced NSCLC using TMB
Radiology [74] Predicting response to immunotherapy in NSCLC AUC, 0.79
using MSI
Pathology [72] Predicting response to immunotherapy in AUC, 0.80
advanced melanoma
Pathology [73] Predicting response to immunotherapy in AUC > 0.99
gastrointestinal cancer using MSI
(Continues)
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1106 CHEN et al.

TA B L E 2 (Continued)

Application Reference Task Performance

Chemotherapy
Radiology [75] Predicting response to NAC in breast cancer AUC, 0.851
Radiology [76] Predicting response to NAC in breast cancer Accuracy, 88%
Radiology [77] Prediction response to NAC in rectal cancer AUC, 0.83
Radiology [78] Prediction response to NAC in NPC Concordance index, 0.719-0.757
Radiology [79] Prediction response to NAC in NPC Concordance index, 0.722
Radiotherapy
Radiotherapy [84] Segmentation of OAR in head and neck DSC, 37.4%-89.5%
Radiotherapy [85] Segmentation of OAR in NPC DSC, 86.1%
Radiotherapy [86] Segmentation of OAR in head and neck DSC, 74%
Radiotherapy [87] Segmentation of OAR in head and neck DSC, 60-83%
Radiotherapy [88] Segmentation of OAR in head and neck DSC, 53-90%
Radiotherapy [91] 3D liver segmentation DSC, 97.25%
Radiotherapy [92] Segmentation of CTV and OAR in rectal cancer CTV: DSC, 87.7%
OAR: DSC, 61.8-93.4%
Radiotherapy [93] Segmentation of OAR in esophageal cancer DSC, 84-97%
Radiotherapy [94] Contouring of GTV in NPC DSC, 79%
Radiotherapy [95] Segmentation of CTV and OAR in cervical cancer CTV: DSC, 86%
OAR: DSC, 82-91%
Radiotherapy [96] Contouring of GTV in colorectal carcinoma DSC, 75.5%
Radiotherapy [97] Contouring of CTV in NSCLC DSC, 75%
Radiotherapy [98] Contouring of CTV in breast cancer DSC, 91%
Radiotherapy [99] IMRT planning in NPC Conformity index, 1.18-1.42
Radiotherapy [102] Prediction of dose distribution of IMRT in NPC Dose difference, 4.7%
Radiotherapy [103] Prediction of three-dimensional dose distribution Dose difference, 2-4.2%
of helical tomotherapy
Radiotherapy [104] Prediction of dose distribution of IMRT in Dose difference, 1.26-5.07%
prostate cancer
Radiotherapy [105] Prediction of three-dimensional dose distribution Dose difference < 0.5%
Abbreviations: AUC, area under curve; NPC, nasopharyngeal carcinoma; MRI, magnetic resonance images; MSI, microsatellite instability; TMB, tumor mutation
burden; NSCLC, non-small cell lung cancer; NAC, neoadjuvant chemotherapy; DSC, Dice similarity coefficient; OAR, organs at risk; GTV, gross tumor volume;
CTV, clinical target volume; IMRT, intensity-modulated radiation therapy.

An emerging area for the early detection of cancers sies increases, we anticipate that DL models will eliminate
is liquid biopsies for circulating tumor DNA (ctDNA) or the need for manual selection and curation of discrimina-
cell-free DNA (cfDNA) obtained via a simple blood test. tory features, as well as allowing for the combination of
These are particularly important for cancer types that cur- multiple data types to enhance early cancer detection.
rently have no effective screening method. In a promis-
ing work, Cohen et al. [26] developed CancerSEEK for the
early detection and prediction of eight cancer types using 2.2 Cancer diagnosis, classification, and
ctDNA. With CancerSEEK, samples are first classified as grading
cancer-positive using a logistic regression model applied
to 16 gene mutations and the expression levels of 8 plasma CNN-based DL models that can accurately diagnose can-
proteins. The cancer type is then predicted using a random cers, classify cancer subtypes, and identify cancer grades
forest classifier, with accuracies ranging from 39% to 84%. using histopathology (e.g., whole slide imaging [WSI])
Although liquid biopsies are promising for early cancer [29], radiology (e.g., CT and magnetic resonance imag-
detection, so far, they have been limited to traditional ML ing [MRI]), and endoscopy images (e.g., esophagogastro-
algorithms [27, 28]. As data acquisition from liquid biop- duodenoscopy and colonoscopy) have been extensively
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1107

reported, and most of them exhibit accuracies at least grades (low versus high) using MRI images with a reported
equivalent to that of professionals. AUC of 0.83. Overall, these studies show the promising
For cancer diagnosis, CNN-based DL models have application of AI in cancer classification and grading, with
exhibited exceptional accuracy in identifying malignant performances equal to trained experts.
tumors using histopathology slides [30–35]. In an inter- From a technical and practical aspect, these DL-based
national competition (CAMELYON16) for diagnosing diagnostic tools integrate features for fine-tuning and
breast cancer metastasis in lymph nodes using WSI with enhancing performance, which simplifies the pipelines of
hematoxylin-eosin (HE) staining, the best CNN algorithm conventional computer-aided diagnosis and reduces false
(a GoogLeNet architecture-based model) yielded an AUC positive rates [54]. Although preclinical assessments of AI
of 0.994, outperforming the best pathologist with an AUC tools have paved the way for clinical trials to improve the
of 0.884 and in a more time-efficient manner [33]. DL accuracy and efficiency of cancer diagnoses, the robustness
algorithms have also been adopted to predict the origin of and generalizability of DL models need to be improved
unknown primary cancers, which is extremely challenging [55].
in cancer diagnosis [36].
The success of DL has also been consistently reported
in the diagnosis of malignant diseases using CT, MRI, 2.3 Predicting gene mutations in cancer
positron emission tomography-CT (PET-CT) scans [37–41],
and endoscopy [42–50]. Most recently, Yuan et al. [40] used DL algorithms have also been used to characterize the
CT scans to develop a classifier using a three-dimensional underlying genetic and epigenetic heterogeneity using
(3D) ResNet algorithm to predict occult peritoneal metas- histopathology images. Using HE-stained WSI of lung can-
tasis in colorectal cancer with an AUC of 0.922, which cer, a CNN was trained to predict six different genetic
was substantially higher than that achieved via routine mutations with an AUC from 0.733 to 0.856 as measured
contrast-enhanced CT diagnosis (AUC = 0.791). In another on a held-out testing cohort [51]. Using WSI, the CNN
work, Ke et al. [41] used MRI images from 4,100 patients model (Inception-V3) also identified common mutations
with nasopharyngeal carcinoma (NPC) to train and test a in liver cancer with AUCs >0.71 [56]. Using WSI, DL tools
self-constrained 3D DenseNet that could distinguish NPC have also been developed for the prediction of whole-
from benign nasopharyngeal hyperplasia with a reported genome duplications, chromosome arm gains and losses,
AUC of 0.95-0.97. As for endoscopy, in a multicenter study, focal amplifications and deletions, and gene variations for
Luo et al. [43] developed a gastrointestinal AI diagnos- pan-cancer [57, 58]. Expanded from predicting mutations
tic system (GRAIDS) for the diagnosis of upper gastroin- in individual genes, DL models have been used to predict
testinal cancers using a CNN-based model and tested it mutational footprints, such as microsatellite instability
in a prospective study involving six different tiered hospi- (MSI) status and tumor mutational burden (TMB) status,
tals. While the diagnostic accuracies varied from 0.915 to which are the most important biomarkers for responses
0.977 among the six hospitals, they were similar to those of to checkpoint immunotherapy. Most recently, Yamashita
expert endoscopists and superior to those of non-experts, et al. [59] trained and tested MSINet, a transfer learning
thus indicating the potential benefit in improving the diag- model based on MobileNetV2 architecture, to classify MSI
nostic effectiveness of community-based hospitals. All in status in HE-stained WSI in a colorectal cancer cohort of
all, such models, if their performance is confirmed in mul- 100 primary tumors and reported an AUC of 0.93. Using
ticenter prospective studies, may play an important role in multiple instances of learning-based DL, Cao et al. [60]
making cancer diagnosis more accurate, especially in local also tried to classify MSI status using WSI in a colorectal
hospitals that lack experts. cancer cohort and achieved an AUC of 0.85. In a work to
Aside from dichotomous diagnosis, DL models are classify TMB status using WSI, Wang et al. [61] compared
used for more challenging cancer classifications and grad- eight different DL models and reported GoogLeNet as the
ing tasks. Coudray et al. [51] developed DeepPATH, an best model for gastric tumors (AUC = 0.75) and VGG-19 as
Inception-v3 architecture-based model, to classify WSI for the best for colon cancer (AUC = 0.82). The results indi-
lung tissues into three classes (normal, lung adenocarci- cate that features from histopathology images can be used
noma, and lung squamous cell carcinoma) with a reported to predict genetic mutations in cases in which obtaining
AUC of 0.97. The CNN was also successfully trained to per- tumor specimens for mutation analysis are not possible.
form automated Gleason grading of prostate adenocarci- Notably, it may be more cost-effective than direct sequenc-
noma, with a 75% agreement between the algorithm and ing.
pathologists [52]. Cancer grading can also be done using In addition to histopathology images, identifying cancer
radiology images. Zhou et al. [53] developed a DL approach mutations using noninvasive radiology images such as
(based on SENet and DenseNet) to predict liver cancer CT or MRI scans has been explored. For example, the
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1108 CHEN et al.

prediction of EGFR mutation status in non-small cell that the classifier accurately stratified patients into respon-
lung cancer (NSCLC) can be achieved using CT and ders and non-responders with an AUC of 0.80. Most excit-
PET/CT scans using DL models both with AUCs >0.81 ingly, Arbour et al. [74] developed a DL model that directly
[62, 63]. In another work, Shboul et al. [64] introduced predicts the best overall response and progression-free sur-
a ML approach to predict O6-methylguanine-DNA vival using radiology text reports for patients with NSCLC
methyltransferase methylation, isocitrate dehydrogenase treated with a programmed cell death protein-1 blockade.
mutation, 1p/19q co-deletion, alpha-thalassemia/mental These studies underscore the potential ability of AI to
retardation syndrome X-linked mutation, and telomerase identify individuals who may benefit from immunother-
reverse transcriptase mutation of low-grade gliomas with apy without the aforementioned negatives of biopsies.
radiomics, and achieved AUCs from 0.70 to 0.84. CT scans In addition to immunotherapy, other therapies (e.g.,
have also been used to predict TMB status in NSCLC (AUC targeted therapy and neoadjuvant chemotherapy [NAC])
= 0.81). The results were promising, but understanding have achieved prominent clinical success in specific pop-
what features are being learned by the CNN models to ulations, driving the need for accurate predictive assays to
determine mutation status remains under researched. inform patient selection. This requirement can be met by a
combination of big data and AI. AI predictive models can
identify imaging phenotypes that are associated with a tar-
3 PATIENT PROGNOSIS, RESPONSE geted mutation. This AI-based approach has the advantage
TO THERAPY, AND PRECISION of identifying the mutation status repeatedly and nonin-
MEDICINE vasively. This approach was supported by a PET/CT-based
DL model for patients with NSCLC, which uses radiomic
Precision medicine refers to the tailoring of treatment to features to discriminate EGFR-mutant types from wild-
individual patients [65]. It aims to classify individuals into type with an AUC of 0.81 [62]. Moreover, with a large
subgroups with differences in their disease prognosis or amount of radiomics data, DL algorithms have shown
in their response to a specific treatment and thus make power in estimating responses to NAC for patients with
therapeutic interventions for those who will benefit and breast cancer [75, 76], rectal cancer [77], and NPC [78, 79].
sparing expense and side effects for those who will not. After NAC, about 35% of patients with locally advanced
DL algorithms are used to automatically extract features breast cancer achieved a pathologic complete response
from medical data to build models that can accurately pre- (pCR), which was associated with improved survival [80].
dict risk of tumor relapse and patients’ responses to treat- Whereas, a poor response to NAC was associated with
ments [66–68]. Based on the prediction results, physicians an adverse prognosis [81]. Therefore, the accurate predic-
can provide more precise and suitable treatments. tion of treatment response is warranted, which can avoid
Immunotherapy drugs have been approved for the treat- unnecessary toxicity and delays to surgery. Using pretreat-
ment of metastatic melanoma, lung cancer, and other ment MRI from patients with locally advanced breast can-
malignancies. However, more than 50%-80% of cancer cer, Ha et al. [76] trained a CNN to predict pCR, and no
patients fail to respond to checkpoint inhibitor therapy. response/progression after NAC, reaching an overall accu-
Currently, response prediction for immunotherapies is racy of 88%. In addition to predicting patient responses to
based on biomarkers of the immunogenic tumor microen- therapies, AI now offers additional avenues to adjust drug
vironment, such as programmed death-ligand 1 (PD-L1) dosage for single or combinational therapies for individual
expression, TMB, MSI, and somatic copy number alter- patients in a dynamic manner using patient-specific data
ations. However, these biomarker data were acquired via collected over time [82].
a biopsy, which is invasive, difficult to perform longitudi-
nally, and limited to a single tumor region. Furthermore,
the predictive value of biomarkers may be limited. In the 4 DEEP LEANING IN RADIOTHERAPY
KEYNOTE-189 clinical trial, immunotherapy with pem-
brolizumab combined with standard chemotherapy pro- Radiotherapy constitutes an integral modality in the treat-
vided survival benefits for all patients regardless of their ment of cancers with half of patients receiving it. The
PD-L1 expression [69]. To achieve the goal of precision image-, data-driven and quality assurance frameworks of
medicine, many researchers have established DL models radiotherapy provide an excellent foundation for the devel-
to predict patient biomarkers related to immunotherapy opment of AI algorithms and their integration into radio-
using radiomics and pathomics data [70–73]. Johannet el therapy workflows. There has been an acute interest in
al. developed a pipeline that integrates DL on histology exploring AI to facilitate radiotherapy for target volume
specimens with clinical data to predict immunotherapy and organs at risk (OAR) delineation and automated treat-
response in advanced melanoma [72]. The results showed ment planning [83].
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1109

Target volume and OAR delineation is a labor-intensive for individualized 3D dose prediction and optimization
process, and its accuracy depends heavily on the experi- [101–104]. Fan et al. [105] first developed an automated
ence of the radiation oncologists. CNN-based semantic seg- treatment planning strategy based on ResNet to achieve an
mentation has been consistently established as a state-of- accurate 3D dose prediction and voxel-by-voxel dose opti-
the-art tool in the automated delineation of OAR in head mization for head and neck cancers. The results showed no
and neck [84–88], thorax [89], abdomen [90, 91] and pelvic significant difference between the predicted and real clin-
regions [92]. OAR is usually delineated on CT images, and ical plans for most clinically relevant dosimetry indices.
the runtime for each patient lasts only several seconds. More importantly, with this strategy, patients with differ-
From these published studies, the segmentation accuracies ent prescription doses can be learned and predicted in a
of organs with large volumes, rigid and regular shapes were single framework.
rather high, such as those of the mandible (Dice similar- Other applications of AI in radiotherapy include
ity coefficient [DSC] = 0.94), parotid (DSC = 0.84), kid- the prediction of radiation-induced toxicities [106–108],
ney (DSC = 0.96), and liver (DSC = 0.97), while for organs image reconstruction [109–111], synthetic CT generation
with small volumes, movable and irregular shapes, the [112–114], image registration [115–117], and intra- and inter-
segmentation accuracies decreased, such as those of the fraction motion monitoring [118–120]. In summary, AI
optic nerve (DSC = 0.69), chiasm (DSC = 0.37), intestine has the potential to improve the accuracy, efficiency and
(DSC = 0.65), and esophagus (DSC = 0.83). Of note, pre- quality of radiotherapy. Furthermore, MRI-only radiother-
liminary studies have shown that differences in dosimetry apy [121] and real-time adaptive radiotherapy [109] could
parameters between automatic and manual delineations be achieved with the implementation of effective and
were small, and automatic segmentations performed suf- efficient automated segmentation, image processing, and
ficiently well for treatment planning purposes [87, 93]. automated treatment planning tools based on DL, which
Given the variety of shapes, locations, and internal mor- are significantly faster than standard approaches.
phologies of tumors, automated contouring of tumor tar-
gets by DL is still a great challenge. Nonetheless, automatic
contouring speeds up the process and improves consis- 5 DL IN CANCER RESEARCH
tency among radiation oncologists. Automated delineation
of the gross tumor volume (GTV) and clinical target vol- DL approaches have been applied in various aspects
ume (CTV) have been investigated in many cancers, such of cancer research, including investigating biological
as nasopharyngeal [94], cervical [95], colorectal [92, 96], underpinnings, developing anti-cancer therapeutics, and
lung [97] and breast cancers [98]. Lin et al. [94] first con- implementing randomized controlled trials (RCTs). To
structed an automated contouring tool for NPC by apply- uncover the biological mechanisms of cancer, studies have
ing a 3D CNN model to MRI. In this independent test, they used DL to analyze the relationship between genotypes
found acceptable concordance between the AI tool and and phenotypes with a large number of achievements
human experts, with an overall accuracy of 79%. Moreover, already reported. In a recent study leveraging DL algo-
in a multicenter test involving eight radiation oncologists rithms, the role of F-box/WD repeat-containing protein 7
from seven hospitals, the AI tool outperformed half of the (Fbw7) in cancer cell oxidative metabolism was discovered
physicians and was equal to the other four. With AI’s assis- via gene expression signatures from The Cancer Gene
tance, substantial improvement in the contouring accu- Atlas dataset [122]. Watson for Genomics also recognized
racy among five of the eight physicians as well as signif- genomic alterations with potential clinical effects that
icant reductions in the interobserver variation (by 54.5%) were not identified by the conventional molecular tumor
and contouring time (by 39.4%) were observed. boards across a spectrum of cancer types [123]. Identifica-
Another important application of AI in radiotherapy is tion of these genetic variants not only pinpoints relevant
automated treatment planning. Radiotherapy planning is biological pathways but also suggests targets for drug
a complex process that involves “trial-and-error” based on discovery. ML methods have also been employed to accel-
physicists’ subjective priorities to achieve specific dosime- erate the early discovery of potential anti-cancer agents
try objectives. As a result, treatment planning quality [124–129]. Valeria et al. [129] reported the first perturbation
depends heavily on the experience of the clinical physi- model combined with ML to enable the design and pre-
cists. While automated planning using knowledge-based diction of dual inhibitors cyclin-dependent kinases 4 and
techniques, such as RapidPlan in Eclipse, have improved human epidermal growth factor receptor 2 with sensitivity
the consistency of planning quality [99, 100], these meth- and specificity higher than 80%. Another key aspect of
ods are suboptimal since they cannot provide estimations drug discovery is the determination of compounds with
of patient-specific achievable dose distributions. Recently, good on-target effects and minimal off-target effects. Zha-
DL-based methods have become a promising approach voronkov et al. [130] developed a DL model and discovered
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1110 CHEN et al.

powerful inhibitors of the discoidin domain receptor 1 To mimic real clinical settings, a multimodal DL model
(DDR1, a kinase target implicated in multiple cancers) in incorporating the aforementioned information plus imag-
just 21 days versus conventional timelines of approximately ing data needs to be constructed in future studies.
one year. All in all, DL AI is accelerating drug discovery
and is already successfully predicting drug behavior.
The adoption of novel cancer treatments is dependent 6.2 Interpretability: the black-box
on successful RCTs. However, successful recruitment of problem
appropriate patients into these trials is regarded as one of
DL has been criticized for being a “black box” that does
the most challenging aspects. Matching complicated eli-
not explain how the model generates outputs from given
gibility criteria to potential subjects is a tedious, labor-
inputs. The large number of parameters involved makes
intensive, and difficult task [131]. To automate this, Has-
it difficult for oncologists to understand how DL models
sanzadeh et al. [132] used natural language processing
analyze data and make decisions. However, some efforts
and a Multi-Layer Perceptron model to extract meaningful
have been made to make this black box more transpar-
information from patient records to help collate evidence
ent [135, 136]. For example, the heat map-like class activa-
for better decision making on the eligibility of patients
tion algorithm, visualizes which image regions are taken
according to certain inclusion and exclusion criteria,. It
into account with DL models when making decisions and
achieved an overall micro-F1 score of 84%. Selecting top-
to what degree. These innovative studies render DL tools
enrolling investigators is also essential for the efficient
execution of RCTs. To facilitate the automation of selec- more interpretable and applicable in clinical oncology set-
tion, Gligorijevic et al. [133] proposed a DL approach to tings.
learn from both investigator- and trial-related heteroge-
neous data sources and rank investigators based on their 6.3 Data access and medical ethics
expected enrollment performance in new RCTs. Here, DL
shows the potential to optimize clinical cancer trials. DL studies not only face technological challenges but also
resource and ethical challenges. The power and believabil-
ity of DL relies on a large amount of training data. Lim-
6 CHALLENGES AND FUTURE ited data may cause overfitting, yielding an inferior per-
IMPLICATIONS formance in an external test cohort [134]. Given the con-
cerns of protecting patient information, medical data are
While AI is widely investigated in oncology, studies need to often the property of individual institutions, and there is
be performed to translate DL models into real-world appli- a lack of data-sharing systems to link institutions. For-
cations. Barriers to improving doctors’ acceptance and per- tunately, this obstacle is beginning to be overcome, with
formance of clinically applied DL include the generaliz- privacy-preserving distributed DL (DDL) and multicen-
ability of its applications, the interpretability of algorithms, ter data-sharing agreements [137–139]. DDL provides a
data access, and medical ethics. privacy-preserving solution to enable multiple parties to
jointly learn via a deep model without explicitly sharing
local datasets. The Cancer Imaging Archive, which col-
6.1 Generalizability and real-world lects clinical images from different institutes and hospitals,
application also provides a good example of data sharing and may pro-
mote radiomic studies [140]. In the future, an authorita-
Because of the great heterogeneity in medical data across tive framework should be developed by governments and
institutions, the performance of DL models tends to enterprises to realize secure data sharing. In addition, sev-
decrease when applied at different hospitals, therefore, eral ethical issues need to be addressed prior to the clinical
external validation sets may be required to confirm their implementation of DL tools. First, the degree of supervi-
performance [55]. Additionally, the extremely large num- sion required from physicians should be determined. Sec-
ber of parameters in DL results in a high likelihood of over- ond, the responsible party for incorrect decisions made by
fitting and limiting of the generalizability across different DL tools should also be determined.
populations [134]. More importantly, in clinical settings,
to make a precise decision, oncologists need to consider
a variety of data, including clinical manifestations, lab- 7 CONCLUSIONS
oratory examinations, imaging data, and epidemiological
histories. However, most recent studies have only adopted DL is a newly developed AI method in oncology which
one type of data (such as imaging) as the input model. is rapidly progressing. With the growth of high-quality
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1111

medical data and the development of algorithms, DL meth- 8. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM,
ods have great potential in improving the precision and et al. Dermatologist-level classification of skin cancer with deep
efficiency of cancer diagnosis and treatment. Moreover, neural networks. Nature. 2017;542(7639):115–8.
9. Krizhevsky A, Sutskever I, Hinton GE. Imagenet classifica-
the positive attitude of the FDA towards AI medical devices
tion with deep convolutional neural networks. Commun ACM.
further increases the prospect of DL’s practical application
2017;60(6):84–90.
in oncology. For the realization of clinical implementation, 10. Bi WL, Hosny A, Schabath MB, Giger ML, Birkbak NJ,
future researches should focus on the reproducibility and Mehrtash A, et al. Artificial intelligence in cancer imag-
interpretability to make DL methods more applicable. ing: Clinical challenges and applications. CA Cancer J Clin.
2019;69(2):127–57.
D E C L A R AT I O N S 11. Aberle DR, Adams AM, Berg CD, Black WC, Clapp JD,
E T H I C S A P P R O VA L A N D C O N S E N T T O Fagerstrom RM, et al. Reduced lung-cancer mortality with
low-dose computed tomographic screening. N Engl J Med.
PA R T I C I PA T E
2011;365(5):395–409.
Not applicable. 12. Byers T, Wender RC, Jemal A, Baskies AM, Ward EE, Brawley
OW. The American Cancer Society challenge goal to reduce US
C O N S E N T F O R P U B L I C AT I O N cancer mortality by 50% between 1990 and 2015: Results and
Not applicable. reflections. CA Cancer J Clin. 2016;66(5):359–69.
13. Corley DA, Jensen CD, Marks AR, Zhao WK, Lee JK, Doubeni
AVA I L A B I L I T Y O F D A T A A N D CA, et al. Adenoma detection rate and risk of colorectal cancer
and death. N Engl J Med. 2014;370(14):1298–306.
M AT E R I A L S
14. Wentzensen N, Lahrmann B, Clarke MA, Kinney W, Tokugawa
Not applicable.
D, Poitras N, et al. Accuracy and efficiency of deep-learning-
based automation of dual stain cytology in cervical cancer
COMPETING INTERESTS screening. J Natl Cancer Inst. 2021;113(1):72–9.
The authors declare that they have no competing interests. 15. Wang P, Berzin TM, Glissen Brown JR, Bharadwaj S, Becq A,
Xiao X, et al. Real-time automatic detection system increases
AU T H O R CO N T R I B U T I O N S colonoscopic polyp and adenoma detection rates: a prospective
ZHC, LL, YS, and RHX conceived this study. ZHC, LL, and randomised controlled study. Gut. 2019;68(10):1813–9.
16. Zhao W, Yang J, Sun Y, Li C, Wu W, Jin L, et al. 3D
CFW drafted the manuscript. YS, RHX, and CFL revised
deep learning from CT scans predicts tumor invasiveness
the manuscript. All authors read and approved the final
of subcentimeter pulmonary adenocarcinomas. Cancer Res.
manuscript. 2018;78(24):6881–9.
17. Kang G, Liu K, Hou B, Zhang N. 3D multi-view convolu-
AC K N OW L E D G M E N T S tional neural networks for lung nodule classification. PLoS
Not applicable. One. 2017;12(11):e0188290.
18. Ciompi F, Chung K, van Riel SJ, Setio AAA, Gerke PK,
ORCID Jacobs C, et al. Towards automatic pulmonary nodule man-
agement in lung cancer screening with deep learning. Sci Rep.
Rui-Hua Xu https://orcid.org/0000-0001-9771-8534
2017;7:46479.
Ying Sun https://orcid.org/0000-0002-5888-2929 19. Ardila D, Kiraly AP, Bharadwaj S, Choi B, Reicher JJ, Peng L,
et al. End-to-end lung cancer screening with three-dimensional
REFERENCES deep learning on low-dose chest computed tomography. Nat
1. McCarthy J, Minsky ML, Rochester N, Shannon CE. A proposal Med. 2019;25(6):954–61.
for the dartmouth summer research project on artificial intelli- 20. Swiderski B, Kurek J, Osowski S, Kruk M, Barhoumi W, edi-
gence, august 31, 1955. AI Magazine. 2006;27(4):12. tors. Deep learning and non-negative matrix factorization in
2. Yasser E-M, Honavar V, Hall A. Artificial Intelligence Research recognition of mammograms. Eighth International Conference
Laboratory. 2005. on Graphic and Image Processing (ICGIP 2016); 2017: Interna-
3. Yu KH, Beam AL, Kohane IS. Artificial intelligence in health- tional Society for Optics and Photonics.
care. Nat Biomed Eng. 2018;2(10):719–31. 21. Arevalo J, González FA, Ramos-Pollán R, Oliveira JL, Lopez
4. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2020. CA Can- MAG, editors. Convolutional neural networks for mammogra-
cer J Clin. 2020;70(1):7–30. phy mass lesion classification. 2015 37th Annual international
5. Wainberg M, Merico D, Delong A, Frey BJ. Deep learning in conference of the IEEE engineering in medicine and biology
biomedicine. Nat Biotechnol. 2018;36(9):829–38. society (EMBC); 2015: IEEE.
6. Meyer P, Noblet V, Mazzara C, Lallement A. Survey on deep 22. Samala RK, Chan HP, Hadjiiski L, Helvie MA, Wei J, Cha K.
learning for radiotherapy. Comput Biol Med. 2018;98:126–46. Mass detection in digital breast tomosynthesis: Deep convolu-
7. Samuel A. Some studies in machine learning using the game of tional neural network with transfer learning from mammogra-
checkers. IBMJ Res Dev. 1959;3:210–29. phy. Med Phys. 2016;43(12):6654–66.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1112 CHEN et al.

23. Huynh BQ, Li H, Giger ML. Digital mammographic tumor clas- 39. Liu S, Zheng H, Feng Y, Li W, editors. Prostate cancer diagno-
sification using transfer learning from deep convolutional neu- sis using deep learning with 3D multiparametric MRI. Medical
ral networks. Journal of Medical Imaging. 2016;3(3):034501. imaging 2017: computer-aided diagnosis; 2017: International
24. Antropova N, Huynh BQ, Giger ML. A deep feature fusion Society for Optics and Photonics.
methodology for breast cancer diagnosis demonstrated 40. Yuan Z, Xu T, Cai J, Zhao Y, Cao W, Fichera A, et al. Develop-
on three imaging modality datasets. Medical Physics. ment and Validation of an Image-based Deep Learning Algo-
2017;44(10):5162–71. rithm for Detection of Synchronous Peritoneal Carcinomatosis
25. McKinney SM, Sieniek M, Godbole V, Godwin J, Antropova N, in Colorectal Cancer. Ann Surg. 2020.
Ashrafian H, et al. International evaluation of an AI system for 41. Ke L, Deng Y, Xia W, Qiang M, Chen X, Liu K, et al. Devel-
breast cancer screening. Nature. 2020;577(7788):89–94. opment of a self-constrained 3D DenseNet model in automatic
26. Cohen JD, Li L, Wang Y, Thoburn C, Afsari B, Danilova L, et al. detection and segmentation of nasopharyngeal carcinoma
Detection and localization of surgically resectable cancers with using magnetic resonance images. Oral Oncol. 2020;110:104862.
a multi-analyte blood test. Science. 2018;359(6378):926–30. 42. Lee JH, Kim YJ, Kim YW, Park S, Choi Y-i, Kim YJ, et al. Spot-
27. Chabon JJ, Hamilton EG, Kurtz DM, Esfahani MS, Mod- ting malignancies from gastric endoscopic images using deep
ing EJ, Stehr H, et al. Integrating genomic features for non- learning. Surg Endosc. 2019;33(11):3790–7.
invasive early lung cancer detection. Nature. 2020;580(7802): 43. Luo H, Xu G, Li C, He L, Luo L, Wang Z, et al. Real-time arti-
245–51. ficial intelligence for detection of upper gastrointestinal can-
28. Mouliere F, Chandrananda D, Piskorz AM, Moore EK, Morris cer by endoscopy: a multicentre, case-control, diagnostic study.
J, Ahlborn LB, et al. Enhanced detection of circulating tumor Lancet Oncol. 2019;20(12):1645–54.
DNA by fragment size analysis. Sci Transl Med. 2018;10(466). 44. Urban G, Tripathi P, Alkayali T, Mittal M, Jalali F, Karnes W,
29. Echle A, Rindtorff NT, Brinker TJ, Luedde T, Pearson AT, et al. Deep learning localizes and identifies polyps in real time
Kather JN. Deep learning in cancer pathology: a new genera- with 96% accuracy in screening colonoscopy. Gastroenterology.
tion of clinical biomarkers. Br J Cancer. 2021;124(4):686-96. 2018;155(4):1069–78. e8.
30. Cruz-Roa A, Gilmore H, Basavanhally A, Feldman M, Gane- 45. Takeda K, Kudo SE, Mori Y, Misawa M, Kudo T, Wakamura K,
san S, Shih N. Accurate and reproducible invasive breast can- et al. Accuracy of diagnosing invasive colorectal cancer using
cer detection in whole-slide images: a deep learning approach computer-aided endocytoscopy. Endoscopy. 2017;49(8):798–
for quantifying tumor extent. Sci Rep. 2017;7:46450. Epub 802.
2017/04/19. https://doi.org/10.1038/srep46450. 46. Chen PJ, Lin MC, Lai MJ, Lin JC, Lu HH, Tseng VS. Accurate
31. Liu Y, Kohlberger T, Norouzi M, Dahl GE, Smith JL, classification of diminutive colorectal polyps using computer-
Mohtashamian A, et al. Artificial intelligence–based breast aided analysis. Gastroenterology. 2018;154(3):568–75.
cancer nodal metastasis detection: Insights into the black box 47. Ito N, Kawahira H, Nakashima H, Uesato M, Miyauchi H, Mat-
for pathologists. Arch Pathol Lab Med. 2019;143(7):859-68. subara H. Endoscopic diagnostic support system for cT1b col-
32. Jannesari M, Habibzadeh M, Aboulkheyr H, Khosravi P, Ele- orectal cancer using deep learning. Oncology. 2019;96(1):44–50.
mento O, Totonchi M, et al., editors. Breast cancer histopatho- 48. Trasolini R, Byrne MF. Artificial intelligence and deep learning
logical image classification: a deep learning approach. for small bowel capsule endoscopy. Dig Endosc. 2020;33:290–7.
2018 IEEE International Conference on Bioinformatics and 49. Li C, Jing B, Ke L, Li B, Xia W, He C, et al. Development and
Biomedicine (BIBM); 2018: IEEE. validation of an endoscopic images-based deep learning model
33. Bejnordi BE, Veta M, Van Diest PJ, Van Ginneken B, Karsse- for detection with nasopharyngeal malignancies. Cancer Com-
meijer N, Litjens G, et al. Diagnostic assessment of deep learn- mun (Lond). 2018;38(1):59.
ing algorithms for detection of lymph node metastases in 50. Wang P, Xiao X, Glissen Brown JR, Berzin TM, Tu M, Xiong F,
women with breast cancer. JAMA. 2017;318(22):2199–210. et al. Development and validation of a deep-learning algorithm
34. Jiang Y, Yang M, Wang S, Li X, Sun Y. Emerging role of deep for the detection of polyps during colonoscopy. Nat Biomed
learning-based artificial intelligence in tumor pathology. Can- Eng. 2018;2(10):741–8.
cer Commun (Lond). 2020;40(4):154–66. 51. Coudray N, Ocampo PS, Sakellaropoulos T, Narula N, Snuderl
35. Song Z, Zou S, Zhou W, Huang Y, Shao L, Yuan J, et al. M, Fenyö D, et al. Classification and mutation prediction from
Clinically applicable histopathological diagnosis system for non-small cell lung cancer histopathology images using deep
gastric cancer detection using deep learning. Nat Commun. learning. Nat Med. 2018;24(10):1559–67.
2020;11(1):4294. 52. Arvaniti E, Fricker KS, Moret M, Rupp N, Hermanns
36. Lu MY, Chen TY, Williamson DFK, Zhao M, Shady M, Lip- T, Fankhauser C, et al. Automated Gleason grading of
kova J, et al. AI-based pathology predicts origins for cancers of prostate cancer tissue microarrays via deep learning. Sci Rep.
unknown primary. Nature. 2021;594(7861):106–10. 2018;8(1):1–11.
37. Mohsen H, El-Dahshan E-SA, El-Horbaty E-SM, Salem A- 53. Zhou Q, Zhou Z, Chen C, Fan G, Chen G, Heng H, et al. Grading
BM. Classification using deep learning neural networks for of hepatocellular carcinoma using 3D SE-DenseNet in dynamic
brain tumors. Future Computing and Informatics Journal. enhanced MR images. Comput Biol Med. 2019;107:47–57.
2018;3(1):68–71. 54. Liu B, Chi W, Li X, Li P, Liang W, Liu H, et al. Evolving
38. Das A, Acharya UR, Panda SS, Sabut S. Deep learning based the pulmonary nodules diagnosis from classical approaches to
liver cancer detection using watershed transform and Gaus- deep learning-aided decision support: three decades’ develop-
sian mixture model techniques. Cognitive Systems Research. ment course and future prospect. J Cancer Res Clin Oncol.
2019;54:165–75. 2020;146(1):153–85.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1113

55. Zech JR, Badgeley MA, Liu M, Costa AB, Titano JJ, Oermann 71. Bao X, Shi R, Zhao T, Wang Y. Immune landscape and a novel
EK. Variable generalization performance of a deep learning immunotherapy-related gene signature associated with clinical
model to detect pneumonia in chest radiographs: A cross- outcome in early-stage lung adenocarcinoma. J Mol Med (Berl).
sectional study. PLoS Med. 2018;15(11):e1002683. 2020;98(6):805–18.
56. Chen M, Zhang B, Topatana W, Cao J, Zhu H, Juengpanich 72. Johannet P, Coudray N, Donnelly DM, Jour G, Illa-Bochaca
S, et al. Classification and mutation prediction based on I, Xia Y, et al. Using machine learning algorithms to pre-
histopathology H&E images in liver cancer using deep learn- dict immunotherapy response in patients with advanced
ing. NPJ Precis Oncol. 2020;4:14. melanoma. Clin Cancer Res. 2021;27(1):131–40.
57. Fu Y, Jung AW, Torne RV, Gonzalez S, Vöhringer H, Shmatko 73. Kather JN, Pearson AT, Halama N, Jäger D, Krause J, Loosen
A, et al. Pan-cancer computational histopathology reveals SH, et al. Deep learning can predict microsatellite instabil-
mutations, tumor composition and prognosis. Nature Cancer. ity directly from histology in gastrointestinal cancer. Nat Med.
2020;1(8):800–10. 2019;25(7):1054–6.
58. Kather JN, Heij LR, Grabsch HI, Loeffler C, Echle A, Muti HS, 74. Arbour KC, Luu AT, Luo J, Rizvi H, Plodkowski AJ, Sakhi M,
et al. Pan-cancer image-based detection of clinically actionable et al. Deep learning to estimate RECIST in patients with NSCLC
genetic alterations. Nat Cancer. 2020;1(8):789–99. treated with PD-1 blockade. Cancer Discov. 2021;11:59–67.
59. Yamashita R, Long J, Longacre T, Peng L, Berry G, Martin B, 75. Huynh BQ, Antropova N, Giger ML, editors. Comparison of
et al. Deep learning model for the prediction of microsatel- breast DCE-MRI contrast time points for predicting response
lite instability in colorectal cancer: a diagnostic study. Lancet to neoadjuvant chemotherapy using deep convolutional neu-
Oncol. 2021;22(1):132–41. ral network features with transfer learning. Medical imaging
60. Cao R, Yang F, Ma SC, Liu L, Zhao Y, Li Y, et al. Development 2017: computer-aided diagnosis; 2017: International Society for
and interpretation of a pathomics-based model for the predic- Optics and Photonics.
tion of microsatellite instability in Colorectal Cancer. Thera- 76. Ha R, Chin C, Karcich J, Liu MZ, Chang P, Mutasa S, et al.
nostics. 2020;10(24):11080–91. Prior to initiation of chemotherapy, can we predict breast
61. Wang L, Yudi J, Qiao Y, Zeng N, Yu R. A novel approach tumor response? Deep learning convolutional neural networks
combined transfer learning and deep learning to predict approach using a breast MRI tumor dataset. J Digit Imaging.
TMB from histology image. Pattern Recognit Lett. 2020;135: 2019;32(5):693–701.
244–8. 77. Shi L, Zhang Y, Nie K, Sun X, Niu T, Yue N, et al. Machine
62. Mu W, Jiang L, Zhang J, Shi Y, Gray JE, Tunali I, et al. Non- learning for prediction of chemoradiation therapy response
invasive decision support for NSCLC treatment using PET/CT in rectal cancer using pre-treatment and mid-radiation multi-
radiomics. Nat Commun. 2020;11(1):5228. parametric MRI. Magn Reson Imaging. 2019;61:33–40.
63. Wang S, Shi J, Ye Z, Dong D, Yu D, Zhou M, et al. Predict- 78. Qiang M, Li C, Sun Y, Sun Y, Ke L, Xie C, et al. A prognos-
ing EGFR mutation status in lung adenocarcinoma on com- tic predictive system based on deep learning for locoregion-
puted tomography image using deep learning. Eur Respir J. ally advanced nasopharyngeal carcinoma. J Natl Cancer Inst.
2019;53(3):1800986. 2021;113:606–15.
64. Shboul ZA, Chen J, K MI. Prediction of Molecular Mutations 79. Peng H, Dong D, Fang M-J, Li L, Tang L-L, Chen L, et al.
in Diffuse Low-Grade Gliomas using MR Imaging Features. Sci Prognostic value of deep learning PET/CT-based radiomics:
Rep. 2020;10(1):3711. potential role for future individual induction chemotherapy
65. Timmerman L. What’s in a name? A lot, when it comes to ‘pre- in advanced nasopharyngeal carcinoma. Clin Cancer Res.
cision medicine’. Xconomy; 2013. 2019;25(14):4271–9.
66. Bychkov D, Linder N, Turkki R, Nordling S, Kovanen PE, Ver- 80. Cortazar P, Zhang L, Untch M, Mehta K, Costantino JP, Wol-
rill C, et al. Deep learning based tissue analysis predicts out- mark N, et al. Pathological complete response and long-term
come in colorectal cancer. Sci Rep. 2018;8(1):3395. clinical benefit in breast cancer: the CTNeoBC pooled analysis.
67. Courtiol P, Maussion C, Moarii M, Pronier E, Pilcer S, Lancet. 2014;384(9938):164–72.
Sefta M, et al. Deep learning-based classification of mesothe- 81. Cortazar P, Geyer CE, Jr. Pathological complete response
lioma improves prediction of patient outcome. Nat Med. in neoadjuvant treatment of breast cancer. Ann Surg Oncol.
2019;25(10):1519–25. 2015;22(5):1441–6.
68. Yu KH, Zhang C, Berry GJ, Altman RB, Ré C, Rubin DL, et al. 82. Blasiak A, Khong J, Kee T. CURATE.AI: Optimizing per-
Predicting non-small cell lung cancer prognosis by fully auto- sonalized medicine with artificial intelligence. SLAS Technol.
mated microscopic pathology image features. Nat Commun. 2020;25(2):95–105.
2016;7:12474. 83. Sahiner B, Pezeshk A, Hadjiiski LM, Wang X, Drukker K, Cha
69. Gandhi L, Rodríguez-Abreu D, Gadgeel S, Esteban E, Felip KH, et al. Deep learning in medical imaging and radiation ther-
E, De Angelis F, et al. Pembrolizumab plus Chemotherapy apy. Med Phys. 2019;46(1):e1–e36.
in Metastatic Non-Small-Cell Lung Cancer. N Engl J Med. 84. Ibragimov B, Xing L. Segmentation of organs-at-risks in head
2018;378(22):2078–92. and neck CT images using convolutional neural networks. Med
70. He B, Di Dong YS, Zhou C, Fang M, Zhu Y, Zhang H, Phys. 2017;44(2):547–57.
et al. Predicting response to immunotherapy in advanced 85. Liang S, Tang F, Huang X, Yang K, Zhong T, Hu R, et al. Deep-
non-small-cell lung cancer using tumor mutational bur- learning-based detection and segmentation of organs at risk in
den radiomic biomarker. J Immunother Cancer. 2020;8(2): nasopharyngeal carcinoma computed tomographic images for
e000550. radiotherapy planning. Eur Radiol. 2019;29(4):1961–7.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1114 CHEN et al.

86. van Dijk LV, Van den Bosch L, Aljabar P, Peressutti D, Both S, 100. Tseng M, Ho F, Leong YH, Wong LC, Tham IW, Cheo T, et al.
R JHMS, et al. Improving automatic delineation for head and Emerging radiotherapy technologies and trends in nasopha-
neck organs at risk by Deep Learning Contouring. Radiother ryngeal cancer. Cancer Commun (Lond). 2020;40(9):395–405.
Oncol. 2020;142:115–23. 101. Wang C, Zhu X, Hong JC, Zheng D. Artificial Intelligence in
87. van Rooij W, Dahele M, Brandao HR, Delaney AR, Slotman Radiotherapy Treatment Planning: Present and Future. Tech-
BJ, Verbakel WF. Deep learning-based delineation of head and nol Cancer Res Treat. 2019;18:1533033819873922.
neck organs at risk: Geometric and dosimetric evaluation. Int J 102. Chen X, Men K, Li Y, Yi J, Dai J. A feasibility study on an auto-
Radiat Oncol Biol Phys. 2019;104(3):677–84. mated method to generate patient-specific dose distributions
88. Brunenberg EJL, Steinseifer IK, van den Bosch S, Kaanders J, for radiotherapy using deep learning. Med Phys. 2019;46(1):56–
Brouwer CL, Gooding MJ, et al. External validation of deep 64.
learning-based contouring of head and neck organs at risk. 103. Liu Z, Fan J, Li M, Yan H, Hu Z, Huang P, et al. A deep learning
Phys Imaging Radiat Oncol. 2020;15:8–15. method for prediction of three-dimensional dose distribution of
89. Lustberg T, van Soest J, Gooding M, Peressutti D, Aljabar P, van helical tomotherapy. Med Phys. 2019;46(5):1972–83.
der Stoep J, et al. Clinical evaluation of atlas and deep learning 104. Nguyen D, Long T, Jia X, Lu W, Gu X, Iqbal Z, et al. A feasibil-
based automatic contouring for lung cancer. Radiother Oncol. ity study for predicting optimal radiation therapy dose distri-
2018;126(2):312–7. butions of prostate cancer patients from patient anatomy using
90. Kline TL, Korfiatis P, Edwards ME, Blais JD, Czerwiec deep learning. Sci Rep. 2019;9(1):1076.
FS, Harris PC, et al. Performance of an artificial multi- 105. Fan J, Wang J, Chen Z, Hu C, Zhang Z, Hu W. Automatic
observer deep neural network for fully automated segmen- treatment planning based on three-dimensional dose distri-
tation of polycystic kidneys. J Digit Imaging. 2017;30(4): bution predicted from deep learning technique. Med Phys.
442–8. 2019;46(1):370–81.
91. Hu P, Wu F, Peng J, Liang P, Kong D. Automatic 3D liver seg- 106. Valdes G, Solberg TD, Heskel M, Ungar L, Simone CB, 2nd.
mentation based on deep learning and globally optimized sur- Using machine learning to predict radiation pneumonitis
face evolution. Phys Med Biol. 2016;61(24):8676. in patients with stage I non-small cell lung cancer treated
92. Men K, Dai J, Li Y. Automatic segmentation of the clinical tar- with stereotactic body radiation therapy. Phys Med Biol.
get volume and organs at risk in the planning CT for rectal 2016;61(16):6105–20.
cancer using deep dilated convolutional neural networks. Med 107. Zhen X, Chen J, Zhong Z, Hrycushko B, Zhou L, Jiang S, et al.
Phys. 2017;44(12):6377–89. Deep convolutional neural network with transfer learning for
93. Zhu J, Chen X, Yang B, Bi N, Zhang T, Men K, et al. Eval- rectum toxicity prediction in cervical cancer radiotherapy: a
uation of Automatic Segmentation Model With Dosimetric feasibility study. Phys Med Biol. 2017;62(21):8246–63.
Metrics for Radiotherapy of Esophageal Cancer. Front Oncol. 108. Ibragimov B, Toesca D, Chang D, Yuan Y, Koong A, Xing
2020;10:564737. L. Development of deep neural network for individualized
94. Lin L, Dou Q, Jin Y-M, Zhou G-Q, Tang Y-Q, Chen W-L, hepatobiliary toxicity prediction after liver SBRT. Med Phys.
et al. Deep learning for automated contouring of primary tumor 2018;45(10):4763–74.
volumes by MRI for nasopharyngeal carcinoma. Radiology. 109. Terpstra ML, Maspero M, d’Agata F, Stemkens B, Intven
2019;291(3):677–86. MPW, Lagendijk JJW, et al. Deep learning-based image recon-
95. Liu Z, Liu X, Guan H, Zhen H, Sun Y, Chen Q, et al. struction and motion estimation from undersampled radial k-
Development and validation of a deep learning algorithm for space for real-time MRI-guided radiotherapy. Phys Med Biol.
auto-delineation of clinical target volume and organs at risk 2020;65(15):155015.
in cervical cancer radiotherapy. Radiother Oncol. 2020;153: 110. Jiang Z, Chen Y, Zhang Y, Ge Y, Yin FF, Ren L. Aug-
172–9. mentation of CBCT Reconstructed From Under-Sampled Pro-
96. Huang YJ, Dou Q, Wang ZX, Liu LZ, Jin Y, Li CF, et al. 3-D jections Using Deep Learning. IEEE Trans Med Imaging.
RoI-Aware U-Net for Accurate and Efficient Colorectal Tumor 2019;38(11):2705-15.
Segmentation. IEEE Trans Cybern. 2020. 111. Madesta F, Sentker T, Gauer T, Werner R. Self-contained deep
97. Bi N, Wang J, Zhang T, Chen X, Xia W, Miao J, et al. learning-based boosting of 4D cone-beam CT reconstruction.
Deep Learning Improved Clinical Target Volume Contour- Med Phys. 2020;47(11):5619–31.
ing Quality and Efficiency for Postoperative Radiation Ther- 112. Maspero M, Savenije MHF, Dinkla AM, Seevinck PR, Intven
apy in Non-small Cell Lung Cancer Front Oncol. 2019;9: MPW, Jurgenliemk-Schulz IM, et al. Dose evaluation of fast
1192. synthetic-CT generation using a generative adversarial net-
98. Men K, Zhang T, Chen X, Chen B, Tang Y, Wang S, et al. Fully work for general pelvis MR-only radiotherapy. Phys Med Biol.
automatic and robust segmentation of the clinical target vol- 2018;63(18):185001.
ume for radiotherapy of breast cancer using big data and deep 113. Liu Y, Lei Y, Wang Y, Wang T, Ren L, Lin L, et al. MRI-based
learning. Phys Med. 2018;50:13–9. treatment planning for proton radiotherapy: dosimetric vali-
99. Chang ATY, Hung AWM, Cheung FWK, Lee MCH, Chan dation of a deep learning-based liver synthetic CT generation
OSH, Philips H, et al. Comparison of Planning Quality and method. Phys Med Biol. 2019;64(14):145015.
Efficiency Between Conventional and Knowledge-based Algo- 114. Bird D, Nix MG, McCallum H, Teo M, Gilbert A, Casanova N,
rithms in Nasopharyngeal Cancer Patients Using Intensity et al. Multicentre, deep learning, synthetic-CT generation for
Modulated Radiation Therapy. Int J Radiat Oncol Biol Phys. ano-rectal MR-only radiotherapy treatment planning. Radio-
2016;95(3):981–90. ther Oncol. 2020;156:23–8.
 25233548, 2021, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cac2.12215, Wiley Online Library on [28/09/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
CHEN et al. 1115

115. Lei Y, Fu Y, Wang T, Liu Y, Patel P, Curran WJ, et al. 4D-CT Hsp90 Inhibitors. ACS Omega. 2018;3(11):14704–16.
deformable image registration using multiscale unsupervised 129. Kleandrova VV, Scotti MT, Scotti L, Speck-Planche A. Multi-
deep learning. Phys Med Biol. 2020;65(8):085003. target Drug Discovery via PTML Modeling: Applications to the
116. Fu Y, Wang T, Lei Y, Patel P, Jani AB, Curran WJ, et al. Design of Virtual Dual Inhibitors of CDK4 and HER2. Curr Top
Deformable MR-CBCT prostate registration using biome- Med Chem. 2021;21(7):661–75.
chanically constrained deep learning networks. Med Phys. 130. Zhavoronkov A, Ivanenkov YA, Aliper A, Veselov MS, Aladin-
2020;48:253–63. skiy VA, Aladinskaya AV, et al. Deep learning enables rapid
117. Duan L, Ni X, Liu Q, Gong L, Yuan G, Li M, et al. Unsu- identification of potent DDR1 kinase inhibitors. Nat Biotech-
pervised learning for deformable registration of thoracic CT nol. 2019;37(9):1038–40.
and cone-beam CT based on multiscale features matching 131. Kadam RA, Borde SU, Madas SA, Salvi SS, Limaye SS. Chal-
with spatially adaptive weighting. Med Phys. 2020;47(11): lenges in recruitment and retention of clinical trial subjects.
5632–47. Perspect Clin Res. 2016;7(3):137–43.
118. Mylonas A, Keall PJ, Booth JT, Shieh CC, Eade T, Poulsen 132. Hassanzadeh H, Karimi S, Nguyen A. Matching patients to
PR, et al. A deep learning framework for automatic detection clinical trials using semantically enriched document represen-
of arbitrarily shaped fiducial markers in intrafraction fluoro- tation. J Biomed Inform. 2020;105:103406.
scopic images. Med Phys. 2019;46(5):2286–97. 133. Gligorijevic J, Gligorijevic D, Pavlovski M, Milkovits E, Glass
119. Kim KH, Park K, Kim H, Jo B, Ahn SH, Kim C, et al. Facial L, Grier K, et al. Optimizing clinical trials recruitment via deep
expression monitoring system for predicting patient’s sudden learning. J Am Med Inform Assoc. 2019;26(11):1195–202.
movement during radiotherapy using deep learning. J Appl 134. Mummadi SR, Al-Zubaidi A, Hahn PY. Overfitting and
Clin Med Phys. 2020;21(8):191–9. Use of Mismatched Cohorts in Deep Learning Models: Pre-
120. Roggen T, Bobic M, Givehchi N, Scheib SG. Deep Learn- ventable Design Limitations. Am J Respir Crit Care Med.
ing model for markerless tracking in spinal SBRT. Phys Med. 2018;198(4):544-5.
2020;74:66–73. 135. Yang JH, Wright SN, Hamblin M, McCloskey D, Alcan-
121. Florkow MC, Guerreiro F, Zijlstra F, Seravalli E, Janssens GO, tar MA, Schrübbers L, et al. A white-box machine learning
Maduro JH, et al. Deep learning-enabled MRI-only photon and approach for revealing antibiotic mechanisms of action. Cell.
proton therapy treatment planning for paediatric abdominal 2019;177(6):1649–61. e9.
tumours. Radiother Oncol. 2020;153:220–7. 136. Samek W, Wiegand T, Müller K-R. Explainable artificial intelli-
122. Davis RJ, Gönen M, Margineantu DH, Handeli S, Swanger gence: Understanding, visualizing and interpreting deep learn-
J, Hoellerbauer P, et al. Pan-cancer transcriptional signatures ing models. arXiv preprint arXiv:170808296. 2017.
predictive of oncogenic mutations reveal that Fbw7 regulates 137. London JW. Cancer Research Data-Sharing Networks. JCO
cancer cell oxidative metabolism. Proc Natl Acad Sci USA. Clin Cancer Inform. 2018;2:1–3.
2018;115(21):5462–7. 138. Ross JS, Waldstreicher J, Bamford S, Berlin JA, Childers
123. Patel NM, Michelini VV, Snell JM, Balu S, Hoyle AP, K, Desai NR, et al. Overview and experience of the YODA
Parker JS, et al. Enhancing Next-Generation Sequencing- Project with clinical trial data sharing after 5 years. Sci Data.
Guided Cancer Care Through Cognitive Computing. Oncolo- 2018;5:180268.
gist. 2018;23(2):179–85. 139. Duan J, Zhou J, Li Y. Privacy-Preserving distributed deep
124. Speck-Planche A, Scotti MT. BET bromodomain inhibitors: learning based on secret sharing. Information Sciences.
fragment-based in silico design using multi-target QSAR mod- 2020;527:108–27.
els. Mol Divers. 2019;23(3):555–72. 140. Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P,
125. Kleandrova VV, Scotti MT, Scotti L, Nayarisseri A, Speck- et al. The Cancer Imaging Archive (TCIA): maintaining and
Planche A. Cell-based multi-target QSAR model for design of operating a public information repository. J Digit Imaging.
virtual versatile inhibitors of liver cancer cell lines. SAR QSAR 2013;26(6):1045–57.
Environ Res. 2020;31(11):815–36.
126. Speck-Planche A. Multicellular Target QSAR Model for Simul-
taneous Prediction and Design of Anti-Pancreatic Cancer
Agents. ACS Omega. 2019;4(2):3122–32. How to cite this article: Chen Z-H, Lin L, Wu
127. Speck-Planche A, Cordeiro MNDS. Fragment-based in sil-
C-F, Li C-F, Xu R-H, Sun Y. Artificial intelligence
ico modeling of multi-target inhibitors against breast cancer-
for assisting cancer diagnosis and treatment in the
related proteins. Mol Divers. 2017;21(3):511–23.
128. Speck-Planche A. Combining Ensemble Learning with a era of precision medicine. Cancer Commun.
Fragment-Based Topological Approach To Generate New 2021;41:1100–1115. https://doi.org/10.1002/cac2.12215
Molecular Diversity in Drug Discovery: In Silico Design of