SSRG International Journal of Computer Science and Engineering Volume 9 Issue 5, 1-4, May 2022

ISSN: 2348 – 8387 / https://doi.org/10.14445/23488387/IJCSE-V9I5P101 © 2022 Seventh Sense Research Group®

Review Article

A Comparative Study on Lung Cancer Detection using

Deep Learning Algorithms
S. Farjana Farvin1, S. Krishna Mohan2
1
Department of Computer Science and Engineering, Anjalai Ammal Mahalingam Engineering College, Thiruvarur, Tamil
Nadu, India
2
Department of Mechanical Engineering E.G.S.Pillay Engineering College, Nagapattinam, Tamil Nadu, India

Received: 10 March 2022 Revised: 22 April 2022 Accepted: 07 May 2022 Published: 28 May 2022

Abstract - Because of its aggressive nature and late diagnosis at advanced stages, lung cancer is one of the major causes of
cancer-related mortality. Early identification of lung cancer is critical for a person's survival, yet it is a difficult challenge to
solve. Cancer identification is crucial for clinical and epidemiologic reasons since it helps to determine subsequent therapy.
This paper reviews the performance of various deep-learning techniques in detecting lung cancer.

Keywords - Convolution neural network, Machine learning, Deep learning, Recurrent neural network, Deep belief neural
network.
features found in a specific region of the feature map
1. Introduction generated by the convolution layer. Convolution is the
Lung Cancer (L.C.) is one of the biggest public cost outcome of applying a filter to input and getting activation.
burdens in the world. Comparing all cancer-related fatalities,
L.C. contributes more than 25 percent. [1]. Several studies 2.2. D.N.N. (Deep Neural Networks)
have been undertaken using deep CNN-based deep learning D.N.N.s are composed of an input, an output, and
architectures for disease diagnosis. CNN-based models are multiple hidden layers in the interior [3]. All layers are
now being investigated for various applications, including made up of neurons modeled after how neurons in the
detecting and categorising lung nodules. Fig. 1 illustrates the brain process data. Each neuron's weight is adjusted
overall structure for lung nodule identification using Deep during the training phase using a method known as
Learning algorithms. gradient descent. Nonlinearities are added between layers
to match the input and output data. As a result, complex
mappings can be modeled using neural networks. After
Input Lung Preprocessing training, the model can classify unseen data. Neural
Scan Images networks can model complex mappings.
Deep
Learning 2.3. R.N.N. (Recurrent Neural Networks)
Algorithms R.N.N. [4] is a form of Artificial N.N. that operates
Healthy
Final on time series or sequential input. R.N.N.s are neural
Diagnosis networks that contain hidden states and can use previous
Cancerous outputs as inputs. Assuming that inputs and outputs are
independent of one another, the output of a recurrent
Fig. 1 Overall structure for lung nodule identification using Deep neural network is dependent on the sequence's previous
Learning Algorithms
components.

2. Deep Learning Algorithms 3. Literature Review

2.1. CNN (Convolutional Neural Network) This section contains a review of the literature. To
CNN is the most extensively used deep learning identify lung cancer, relevant material from diverse
technique in medical image analysis [2]. The neural sources was referred to.
networks consist of three layers: input, hidden, and output.
The input layer fetches inputs, while the output layer Nasrullah et al. [5] have introduced a based on
produces the outputs. The hidden layer contains the distinctive deep learning methods with different techniques
convolution and pooling layers. The pooling layer has used for detecting malignant lung nodules with high accuracy.
to reduce feature map dimensions and summarize the Faster R-CNN with capable knowledge of CMixNet and U-

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

 S. Farjana Farvin & S. Krishna Mohan / IJCSE, 9(5), 1-4, 2022

Net features, such as encoder-decoder architecture, was used trees and shows lung cancer malignancy predictions as an
to detect nodules. for lung cancer calcification, three types interpretable decision tree. The best decision tree model
of D.N.N. (e.g., CNN, D.N.N., and S.A.E.) have been achieved 86.7 percent sensitivity, 80.0 percent specificity,
developed [3]. These networks are used to classify C.T. and a 92.9 percent positive predictive value.
images, with minor modifications for benign and malignant
lung nodules. Manoj Senthil Kailasam and Meera Devi
Thiagarajan[13] discussed a method for deconstructing sub-
Lakshmi et al. [6] suggested a technique that uses bands using a “Co-active Adaptive Neural Fuzzy Inference
transfer learning to detect lung cancer tissues without the System” (CANFIS) that computes and classifies co-
need to provide features for image classification. With 93 % occurrence features, patterns, and DWT.
and 90 % accuracy, the authors produced results for
Modified VGG16 and Modified VGG19 using a transfer O. Obulesu et al. [14] suggested the Wilcoxon Signed
learning approach. Nikitha Johnsirani Venkatesan et al. [7] Generative Deep Learning (WS-GDL) approach for
proposed an ideal Convolutional Neural Network model diagnosing lung cancer that combines WS-Rank Gain
with decreased Gaussian noise for higher classification and Preprocessing with Generative Deep Learning. To begin, use
training accuracy. test impact examination and information improvement to
reduce duplicate and unnecessary traits while extracting
Huseyin Polat et al. [8] investigated two different many useful and important ones. Using a generator function,
models: Straight 3D-CNN with traditional softmax and the Generative Deep Learning approach was then used to
hybrid 3D-CNN with Radial Basis Function (RBF)-based learn the deep features. Finally, a min-max game (i.e.,
SVM. in the diagnosis of lung cancer; the authors found that minimizing error with high precision) was presented to
a hybrid 3D-CNN with SVM produced better results, with diagnose the disease.
accuracy rates of 91.81 %, 88.53 %, and 91.91 %,
respectively, compared with a straight 3D-CNN softmax. Keerti Maithil et[17] discussed the method for image
segmentation using Dense U-Net+. Using encoding and
According to Xinrong Lu et al. [9], the Marine decoding structure, U-Net deep learning algorithm works for
Predators Algorithm-based method gives the maximum the semantic segmentation of images.
efficiency with the least error, 93.4 %, 98.4 % sensitivity,
and 97.1 % specificity. The authors proposed a new CNN 4. Performance Metrics
for early lung cancer detection. The marine predator's Different criteria are utilized to evaluate the performance
method is also employed for greater network accuracy and of the designed deep-learning algorithms for identifying and
optimal organization. categorizing lung nodules. The authors used measurements
such as specificity (S.P.), sensitivity (S.E.), precision (PPV),
Giang Son Tran et al. [10] discussed an auto feature accuracy (A.C.C.), F1-score, Receiver Operating
extraction and categorization technique of lung candidates as Characteristic (R.O.C.) curve, and area under the R.O.C.
nodule or non-nodule using a unique 15-layer 2D deep CNN curve (A.U.C.) in the reviewed publications [15]. Table 1
architecture. The training process is then enhanced using a depicts the performance metrics for D.N.N. approaches.
focal loss function to improve classification accuracy.

Masud M. et al. [11] used histopathological scans to 5. Performance Evaluation

create a categorization framework for distinguishing In this paper, Table 2 compares different deep-
between five types of lung and colon tissues (two benign and learning models to detect lung cancer with Accuracy,
three malignant). According to the results, the suggested Sensitivity, and Specificity as a percentage [16].
framework can detect cancer tissues up to 96.33 percent of
the time. The performance comparison of various deep learning
algorithms has been analyzed with performance metrics such
Michael Horry et al. [12] developed a novel computer as model accuracy, sensitivity, and specificity, as shown in
vision approach that combines deep learning and decision Fig. 2.

2
 S. Farjana Farvin & S. Krishna Mohan / IJCSE, 9(5), 1-4, 2022

Table 1. Performance metrics for deep learning method

Metric Definition
the number of correct predictions made is a ratio of all predictions made.
Accuracy
(TP + TN)∕n
the proportion of benign lesions successfully diagnosed was determined using the equation:
Specificity
T.N.∕(T.N. + F.P.)
the proportion of malignant lesions is accurately classified. Computed using the following
Sensitivity
equation: T.P.∕(T.P. + F.N.)

Receiver operating
A figure depicting sensitivity versus specificity at various categorization limits (the higher the
characteristic (R.O.C.)
curve, the better the model distinguishes between benign and malignant tumors).
curve

This value represents the area under the R.O.C. curve.

The area under the curve
(A.U.C. = 1.0 implies that the model is perfectly capable of distinguishing between benign and
(A.U.C.)
malignant tumors.)
F1 score is the weighted average of the precision and recall
F1-Score
F1=2TP/(2TP+FN+FP)

Table 2. Different Deep Learning Algorithms and percentage of Accuracy, Sensitivity, and Specificity
Performance Metrics (%)
Name of the Authors Sensitive
Algorithm Occur acy Specific city
vity
84.15 83.96 84.32
QingZeng Song et. al.[3] CNN DNN SAE 82.37 80.66 83.9
82.59 83.96 81.35
Nasrullah Nasrullah, et. al. 2 Deep 3D customized mixed link network
88.79 94 91
[5] (CMixNet)
Unique 15 layers 2D deep CNN(Ldc
Giang Son Tran et al. [10] 97.2 96.0 97.3
Net)
Modified VGG16 93 98 88
D. Lakshmi et al. [6]
Modified VGG19 90 80 100
Hybrid
3D-CNN
Huseyin Polat and Homay
with RBF based
Danaei Mehr[8] 91.91 88.53 94.23
SVM
Hybrid Deep Learning & Decision tree-based
Computer Vision 85 86.7 80.0
Horry, M.; et al. [12]
Model
Co-active adaptive neural fuzzy inference
System
Manoj Senthil (CA 99.2 92.5 98.1
et al. [13] NFIS)
Nikitha Johnsirani NG-CNN
Venkatesan , et. al [7] Algorithm 95 91.39 90.32
al. [9]
Xinrong Lu
M.P.A. 93.4 98.4 97.1
, et
Based Method

3
 S. Farjana Farvin & S. Krishna Mohan / IJCSE, 9(5), 1-4, 2022

6. Conclusion
In this work, we presented different deep-learning
algorithms for Lung cancer detection. Different ways of
evaluation for these methodologies were examined, along
with their relative importance. According to research, Deep
learning approaches have resulted in a good performance for
lung cancer diagnosis using C.T. images. Deep CNN and
Hybrid deep learning techniques were applied in several
proposed systems. It is worthwhile to create a new CAD
system employing U- Net, Faster R-CNN, Mask R-CNN,
YOLO, V.G.G., ResNet, etc. Researchers believe that more
datasets and more balanced data can result in better
Fig. 2 depicts the model accuracy, sensitivity, and specificity of outcomes.
several Deep Learning methods

References
[1] Mehmet A. Gulum, Christopher M. Trombley, and Mehmed Kantardzic, “A Review of Explainable Deep Learning Cancer Detection
Models in Medical Imaging,” Applied Science, vol. 11, no. 10, pp. 1-21, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[2] Sunyi Zheng et al., “Automatic Pulmonary Nodule Detection in C.T. Scans using Convolutional Neural Networks Based on Maximum
Intensity Projection,” IEEE Transactions on Medical Imaging, vol. 39, no. 3, pp. 797–805, 2020. [CrossRef] [Google Scholar]
[Publisher Link]
[3] Qing Zeng Song et al., “Using Deep Learning for Classification of Lung Nodules on Computed Tomography Images,” Journal of
Healthcare Engineering, 2017. [CrossRef] [Google Scholar] [Publisher Link]
[4] Dipanjan Moitra, and Rakesh Kr Mandal, “Automated AJCC (7th Edition) Staging of Non-Small Cell Lung Cancer (NSCLC) Using
Deep Convolutional Neural Network (CNN) and Recurrent Neural Network (R.N.N),” Health Information Science and Systems, vol. 7,
no. 1, pp. 14, 2019. [CrossRef] [Google Scholar] [Publisher Link]
[5] Nasrullah Nasrullah et al., “Automated Lung Nodule Detection and Classification using Deep Learning Combined with Multiple
Strategies,” Sensors, vol. 19, no. 17, pp. 3722, 2019. [CrossRef] [Google Scholar] [Publisher Link]
[6] D. Lakshmi, K. Palani Thanaraj, and M. Arunmozhi, “Convolutional Neural Network in Detecting Lung Carcinoma Using Transfer
Learning Approach,” International Journal of Imaging Systems and Technology, vol. 30, no. 2, pp. 445–454, 2020. [CrossRef] [Google
Scholar] [Publisher Link]
[7] Nikitha Johnsirani Venkatesan, Dong Ryeol Shin, and Dong Ryeol Shin, “Nodule Detection with the Convolutional Neural Network
using Apache Spark and GPU Frameworks,” Applied Science, vol. 11, no. 6, pp. 1-18, 2021. [CrossRef] [Google Scholar] [Publisher
Link]
[8] Huseyin Polat, and Homay Danaei Mehr, “Classification of Pulmonary C.T. Images Using Hybrid 3d-Deep Convolutional Neural
Network Architecture,” Applied Science, vol. 9, no. 5, 2019. [CrossRef] [Google Scholar] [Publisher Link]
[9] Xinrong Lu, Y. A. Nanehkaran, and Maryam Karimi Fard, “A Method for Optimal Detection of Lung Cancer Based on Deep Learning
Optimized by Marine Predators Algorithm,” Computational Intelligence and Neuroscience, vol. 2021. [CrossRef] [Google Scholar]
[Publisher Link]
[10] Giang Son Tran et al., “Improving Accuracy of Lung Nodule Classification using Deep Learning with Focal Loss,” Journal of
Healthcare Engineering, vol. 2019, 2019. [CrossRef] [Google Scholar] [Publisher Link]
[11] Mehedi Masud et al., “A Machine Learning Approach to Diagnosing Lung and Colon Cancer using Deep Learning‐ Based Classification
Framework,” Sensors (Switzerland), vol. 21, no. 3, pp. 1–21, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[12] Michael Horry et al, “Deep Mining Generation of Lung Cancer Malignancy Models From Chest X-Ray Images,” Sensors, vol. 21, no.
19, pp. 1–23, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[13] Manoj Senthil Kailasam, and MeeraDevi Thiagarajan, “Detection of Lung Tumor using Dual-Tree Complex Wavelet Transform and
Co-Active Adaptive Neuro-Fuzzy Inference System Classification Approach,” International Journal of Imaging Systems and
Technology, vol. 31, no. 4, pp. 2032–2046, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[14] O. Obulesu et al, “Adaptive Diagnosis of Lung Cancer by Deep Learning Classification using Wilcoxon Gain and Generator,” Journal of
Healthcare Engineering, vol. 2021, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[15] Diego Riquelme and Moulay Akhloufi, “Deep Learning for Lung Cancer Nodules Detection and Classification in C.T. Scans,” AI, vol. 1,
no. 1, pp. 28–67, 2020. [CrossRef] [Google Scholar] [Publisher Link]
[16] Suren Makaju et al., “Lung Cancer Detection using C.T. Scan Images,” Procedia Computer Science, vol. 125, pp. 107–114, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[17] Keerti Maithil, and Tasneem Bano Rehman, “Urban Remote Sensing Image Segmentation using Dense U-Net+,” SSRG International
Journal of Computer Science and Engineering, vol. 9, no. 3, pp. 21-28, 2022. [CrossRef] [Publisher Link]