A Comprehensive Study of Medical Image Analysis

Gitanjali Ganpatrao Nikam1, Shuchi Singh2, Priya Singh3, Radhika Sarraf4

Department of Computer Applications, National Institute of Technology
Kurukshetra, India
[email protected], [email protected], [email protected], [email protected]
Email:

Abstract— Modern medicine has become reliant on medical the relationship between some functional abnormalities and
imaging. The application of computer has proved to be an diseases. Medical image analysis can be applied for CAD in
emerging technique in medical imaging and medical image mammography, tumor treatment, dental treatment, and
analysis. Each level of analysis requires an effective algorithm dentistry.
as well as methods in order to generate an accurate and
The most common technique of image analysis is image
reliable result. Different modalities, such as X-Ray, Magnetic
resonance imaging (MRI), Ultrasound, Computed tomography
registration. A lot of research has been performed on the
(CT), etc. are used for both diagnoses as well as therapeutic image. Image registration can also be named as image fusion
purposes in which it provides as much information about the or image matching. Image registration can be defined as the
patient as possible. Medical image processing includes image method of transforming various collections of data into one
fusion, matching or warping which is the task of image coordinate system. Data can be from different sensors,
registration. Medical image analysis includes Image multiple photographs, times or viewpoints as presented in [3]
enhancement, segmentation, quantification, registration, which and [23]. For co-registration of medical images, rigid and
is the most predominant ways to analyze the image. There are non-rigid methods are used. Rigid methods often provide
various difficulties in medical image processing and subsequent enough freedom, while non-rigid methods provide elastic
stages like image enhancement and its restoration; transformation and include radial basis function. Non-rigid
segmentation of features; registration and fusion of image registration is an optimization problem where the aim
multimodality images; classification of medical images; image is to optimize an image similarity measure with respect to the
features measurement and analysis and assessment of transformation parameter shown in [19]. Applications of
measurement and development of integrated medical imaging image registration in the medical field include surgery
systems for the medical field. In this paper, the techniques used simulation, radiation therapy, assisted and guided surgery.
in medical image analysis have been reviewed and discussed All these applications have highly relied on the automatic
extensively. The vital goal of this review is to identify the analysis of images. Medical image analysis helps the surgeon
current state of the art of medical image analysis methods as a
to operate or diagnose more reliably. Through registration,
reference paradigm in order to accelerate the performance of
existing methods.
we can integrate the information from various images into a
normalized framework. However, registration-based medical
Keywords— Medical Imaging, Image segmentation, Image image analysis is not trivial to achieve.
Enhancement, Restoration, Registration. The methods employed for image analysis and findings
extracted are summarized in section 2; the contributions of
I. INTRODUCTION each technique and the gaps present in knowledge are
Medical Imaging is the science of solving or analyzing presented and discussed in section 3; comparative study and
medical problems which are based on different medical discussion is furnished in section 3; the final conclusion is
imaging methods and digital image processing techniques in section 4.
[11]. Using these techniques, one can produce images of the II. RELATED WORK AND BACKGROUND
human body or any part of its organ for medical science
purposes viz. Diagnosis, treatment or surgery [14]. Many We are providing a brief description of techniques
computer boffins are trying their best to develop such which will be explained thoroughly in the coming sections.
algorithms and systems in favor of medical experts that can Image Enhancement: It can be defined as the act of
help them to improve the quality of human health. Creating enhancing the perception of information from digital images
well suited and reliable algorithms for analysis of images is or improving the interpretability of an image. It is a process
the major challenge which compels the medical image of making images more appropriate than the previous image
analysis fraternity to work with great endeavors. for a specific application. Image segmentation: It is a
Major tasks in medical image analysis involve image process of segregating a digital image into several regions
enhancement, segmentation, registration of images, (sets of pixels). The goal is to represent an image into a
quantification and computer-aided detection (CAD) meaningful form which is easier to analyze. Image
techniques. Medical image analysis helps a surgeon to registration: Image registration is a method to accomplish
immediately plan the most favorable approach to some target mapping between two different images of mono or
structure as medical imaging are analyzed to organize multimodal images. It is a method to Superimposes the pixel
anatomic structures. Image analysis also helps in inspecting from a reference image to the target image by aligning the

978-1-7281-3250-1/19/$31.00 ©2019 IEEE ©

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on August 07,2020 at 10:42:46 UTC from IEEE Xplore. Restrictions apply.
 image into a common coordinate system. Quantification: B. Image Segmentation
Quantification is measuring the features on medical images, It is also a crucial step in the image analysis approach. It
e.g. helping a radiologist obtain measurements from medical can be defined as the process of discriminating several
images (e.g., area or volume). Computer-aided Detection:
features represented in an image. This info can be easily
Computer-aided diagnosis or detection is a technology
built by edges, contours, medical axes, point sets, areas, etc.
which is designed for decreasing observed oversights and
incorrect negative rate for interpreting medical images by A vital problem in medical image segmentation is a lot of
physicians. unevenness in images. Using segmentation, we spot and
arrange individual region in image viz. Gray matter or white
A. Image Enhancement matter. After this, we get a region of interest (ROI), the
information in which will be used in further stride. Fig. 2
Image enhancement technique plays a very crucial part in shows the original, as well as segmented MRI brain image,
image analysis. Enhancement of little information and was done using adaptive thresholding [27]. Algorithms such
ignoring of other information, image, the visual effect can as intensity measure or target recognition are used to
be improved by it. Image enhancement is a process which calculate the object’s location as presented in [25]. Many
can be used for reducing image noise, removing artifacts image segmentation methods are there which can be divided
and also preserving details. The aim of image enhancement as border based or intensity based as in [9]. Segmentation
is to amplify a few features of an image for processing, suffers from the problem of poor image qualities which are
display, and diagnosis. It can be done by either increasing generated by the situations such as noise, less contrast, the
the contrast of images or suppressing its noise. It can be similarity in the region or parts of the body also very few
defined as the process of adjusting an image so that the scanning methods results in unsuccessful segmentation.
digital image becomes more suitable for subsequent image
analysis. As perception and interpretation are improved, the In the work presented in [2], the threshold based
prominent features can be easily discovered. Also, good clustering and active contour identification has been
quality of an image is essential to perform analysis and adopted for effective segmentation of brain MRI images.
processing stages. However, there is no ideal measure to The analysis helps in Surgery simulation, multiple sclerosis
define a ‘good’ quality of an image. When such a pre- lesion, etc. However, no standard method has been found
processed images are used with the correct techniques, they for the segmentation of all kinds of MRI images.
will give faultless results after analysis. There are two Segmentation is also crucial for a digital map which is made
approaches to image enhancement: up of various map entities such as map symbols, text and
numbers and has to be separated from the background. The
1) By working directly with pixels (Spatial domain approach presented in [12] is based on a shape analysis. The
methods) work presented in [13] uses chain code representation of the
external contour of a map entity. But segmentation of
2) By working on the Fourier transform of the image
(frequency domain methods). medical images has much complications in the process due
to a huge range in topologies, the complexity of medical
Fig.1 shows histogram equalization done on MRI brain structures and low image qualities because of some
image to enhance the quality of the image for further surroundings such as little divergence, noise, and likeness of
processing [26]. body tissues, certain kind of artifacts and restriction of
In the work presented in [1-2], the Histogram equalization scanning method, which outcomes in unsuccessful
and contrast stretching, have been applied to enhance dental segmentation.
x-ray as well as wrist/knee MRI images. However, an
adaptive weighted mean filter (AWMF) failed to remove the
salt and paper noise. It shows limited performance for
severely degraded images. Hence, a more efficient approach
is needed.

Fig. 2 Segmentation of brain MRI image [27]

Fig.1 Enhancement of MRI brain image [26]

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on August 07,2020 at 10:42:46 UTC from IEEE Xplore. Restrictions apply.
 C. Image Registration linear transformations have an advantage over linear
Image registration is a method to accomplish mapping transformations because it has the facility of registering
between two different images of mono or multimodal local deformations also, thus it assures accuracy.
images.
Optimization algorithms. An optimization procedure is
It is a method to Superimposes the pixel from a reference the last element of the image registration process. It selects
image to the target image by aligning the image into a the optimal search techniques to get a transform operator by
common coordinate system. The possible application areas utilizing the results from similarity detection [17].
of registration such as in medical (monitoring of tumor
Continuous Variables: - using a continuous variable the
evaluation, magnetic resonance imaging, MRI, ultrasound,
differential of an objective function is solved by defining
magnetic resonance spectroscopy, specimen classification) several parameters.
and in computer vision (shape recovery, automatic change Discrete Variables: - Unlike continuous variable based
detection, motion tracking, automatic quality optimization the discrete optimization is limited or restricted
inspection)[16]. Images are aligned using methods based on to discrete values only. It uses a Markov random field to
geometrical transformation, point, surface, and intensity. perform optimization
Medical image registration is mainly categorized into
three fields’ viz. Similarity measures, geometrical
transformation, and optimization procedures.
Similarity Measures. Similarity measures can be
categorized into two types, Feature based and the intensity
based methods [22].
Intensity-based similarity measures: Also termed as
template matching or area based matching. It somewhat
merges the feature detection and matching steps. This
concern with the whole image instead of salient points or
features detected in the first step. These methods are
considered an area of the image as features. Intensity-based
similarity measures are commonly based on intensity
differences, information theory and cross-correlation of
intensity. Intensity measures which come under intensity
differences are based on normalization or sum of squared
differences (SSD). Derived measures of cross-correlation -
correlation ratio or person’s correlation coefficient are used Fig. 3 Registration result in images with missing data [4]
as similarity measures. For better image registration the
cross-correlation should be larger. Similarity measures Fig.3 shows registration result with a source, target
under information theory are majorly based on derived unregistered, and registered images. Registration result was
measures or mutual information. obtained after rotation and translation. In the research work
presented in [4], rigid registration has been used for
Feature-based similarity measures: Unlike intensity- Multiphase contrast-enhanced abdomen CT image, chest
based methods which consider an area of the image as a image registration. However, this method is highly restricted
feature these methods are concerned with control points. for orthopedic and brain tumor [10]. Also, very little
Control points are the features which are detected in the research has been done in 3D non-rigid image registration.
feature detection step from a sensed image as well as
reference images [18]. The overall focus is on spatial
relations or various descriptors of features. The feature- D. Quantification
based similarity measure is mostly calculated as the sum of Quantification is measuring the features on medical
distances related to each co-relation build [20]. images, e.g. helping a radiologist obtain measurements from
medical images (e.g., area or volume). For extracting the
Geometric transformations. Geometric transformations essential information like motion, shape, angle, texture, size
correlate with those parameters which need to be optimized of diagnosis quantification algorithm applies to segments as
in parameter optimization as in [17]. Its choice is important in [21]. Fig.4 shows the result of endocardia longitudinal
for successful image registration and is deeply dependent on strain done on the patient without coronary artery disease.
the description of data. Divided into two classes: Quantification’s effectiveness mainly depends on image
fusion and selection of database where different kinds of
Linear Transformation: - It is basically used in a tissue are classified. Quantification of cardiac function, a
condition where the inner structure of the image does not technique for musculoskeletal biomechanics, mammography
show any apparent distortions of deformations. It is are some application of quantification. Quantification
performed by calculating translational and rotational accuracy needs to be improved for a wider range of images.
vectors. The main disadvantage is the lack of a solution to The research work carried out by Bankman [5], illustrates
nearby deformations. the performance of quantification for musculoskeletal
biomechanics and its image analysis in mammography.
Non-linear Transformation: - In linear transformations, However, Quantification accuracy is needed to be improved
we face the problem of local deformations in soft tissues for a wide range of an image.
which affect the accuracy of linear transformations. Non-

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on August 07,2020 at 10:42:46 UTC from IEEE Xplore. Restrictions apply.
 Table 1. Comparative analysis of Medical Image Analysis techniques.

Techniques Methods Limitations Applications Source

Enhancement Histogram Problems Dental X- Abdolvahab
Equalizatio in ray, knee Ehsani Rad,
n, enhancing MRI Alireza
Contrast degraded Norouzi
stretching, images. “Principle
AWMF technique in
medical
analysis
process”
Segmentation Thresholdi- No Surgery Abdolvahab
ng standard simulation, Ehsani Rad,
clustering, segmentati Multiple Alireza
Fig. 4 Endocardia Longitudinal Strain Study to detect a Coronary Artery active on sclerosis Norouzi
Disease [6] contour technique Lesion,etc “Principle
for technique in
satisfactory medical
E. Computer aided Detection result. analysis
Computer aided diagnosis or detection is a technology process”
which is designed for decreasing observed oversights and Registration Rigid, non Very little Multiphase Mohamed
-rigid, research in contrast Abdel
incorrect negative rate for interpreting medical images by affine 3D non enhanced Basset, Arun
physicians. With the help of X-ray, ultrasound, MRI rigid abdomen Kumar,Ahm
medical professional can easily analyze and evaluate registration CT image, edE Fakhry
medical images in very less time period. CAD is mainly chest image “Feature and
registration intensity
used for easily detecting the disease by reduction of false based
negative rate, which is produced due to oversights. Some of medical
the applications of computer-aided diagnosis or detection of image
registration
chest radiography, lung cancer, and breast cancer detection. using PSO”.
CAD, mainly depend on equal roles of the computer and Quantification Spectropho Quantificat Musculoskel Bankman IN
physician, so it is not the same as that of automated t geometric ion etal Handbook of
method accuracy biomechanic medical
computer diagnosis as shown in [24]. Fig. 5 shows typical need to be s, image image
fundus images of normal (a) and different stages of diabetic improve analysis in processing
retinopathy having mild NDPR (b), moderate NDPR (c), for wider mammograp and analysis.
range of hy
severe NDPR (d), prolific DR (e) and macular edema (f) [7]. image
CAD Digital Can’t Chest Kunoi Doi
mammogra detect radiography, “Computer
m, 100% of detection of aided
computer pathologica lung cancer, diagnosis in
aided l changes. breast medical
cytologic cancer imaging
detection. :Historical
review,
current
status, and
future
potential”
.
Image Enhancement can be used in dental X-ray or
knee MRI. The problem in enhancement is the defining
process for discarding degradations. Segmentation can be
Fig.5. Computer-aided diagnosis of diabetic retinopathy [7]
done using clustering and thresholding. However, no ideal
segmentation technique has been found. Image registration
In future CAD can be utilized as an important tool for can be done in two ways using either rigid or non-rigid
diagnostic examination. In [8], Digital mammogram and approach, but very few researches in 3D non-rigid image
Computer-aided Cytology have been used for chest registration are done so far. Quantification has accuracy
radiography, lung cancer detection and breast cancer problems for a wide range of images. CAD systems pursue
detection [15] CAD is unable to detect 100% of pathological to highlight uncertain structures. Today's CAD systems
changes in many cases. failed to show 100% accuracy in detecting pathological
changes. The basic techniques of medical image analysis
III. COMPARATIVE ANALYSIS have been studied and analyzed with their corresponding
performance. In this study, it has been found that mutual
A comparative analysis of medical image analysis image data and its analysis depend upon the object
techniques is provided in Table 1. characteristics and viewing conditions. Hence techniques
must be able to do multi-view analysis of the medical
image. A feature matching approach which is usually
adopted by Image registration has the potential to find
intersecting points. Hence mutual information based

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on August 07,2020 at 10:42:46 UTC from IEEE Xplore. Restrictions apply.
 registration can be an effective way for medical image [10] Stefan Bauerz, Lutz-P. Nolte, Roland Wiest and Mauricio Reyes: “A
analysis as it is data independent. However, very limited survey of MRI-based medical image analysis for brain tumor studies”,
Phys. Med. Biol. 58(13): R97–R129, (2013).
problems with image registration have been addressed yet
[11] A.P. Dhawan, D S Kim and H. K. Huang: “Advanced methods in
by the researchers. medical imaging and image analysis”, 1st ed. Singapore: World
Scientific, (2008)
IV. CONCLUSIONS
[12] Gitanjali G. Nikam, J. K. Ghosh: “Indian topographical map symbols
The fundamental tasks of medical image analysis such as understanding System”, IET Computer Vision, The Institution of
enhancement, segmentation, and registration have been Engineering and Technology, 9(2), 300–308, (2015).
discussed. In this paper, we discussed the concept of [13] Gitanjali G. Nikam, J. K. Ghosh: “A map legend understanding
system, advances in intelligent and soft computing”, 131, 53-62,
principle steps in medical image analysis and described (2012).
some methods. An approach for analyzing medical images [14] Angelini, E. D., E., Konukoglu, E., Clatz, O., Mandonnet Capelle, L.,
leading to variability in multi-view analysis needs to be and Duau, H. Glioma: “Dynamics and computational models: A
developed. As medical imaging is a vital source of the review of segmentation, registration and in Silico growth algorithms
and their clinical applications”. Current Medical Imaging.Reviews,
information of patients, its automatic analysis will benefit 3(4):262-276, (2007)
currently investigating various techniques used in medical [15] Giger ML.: “Computerized analysis of images in the detection and
image analysis. However, stereoscopic visualization of 3D diagnosis of breast cancer. seminars in ultrasound CT and MRI,” 25,
medical imaging is often difficult to render with traditional 411–418 (2004).
image processing techniques. This is not strange that one [16] Barbara Zitova, and Jan Flusser: “Image registration methods: a
could obtain an image with the integration of information survey, Image and Vision Computing”, 21, 977-1000, (2003).
from various sources and start to analyze it. This explains [17] Guoli Song, Yiwen Zhao, Zheng Wang, Jianda Han, and Huibin Du:
“A review on medical image registration as an optimization problem”,
why image registration is one of the most promising Current Medical Imaging Reviews, 13, 274-283, (2017).
approaches towards automated medical image analysis. The [18] Paresh M. Patel and Vishal M. Shah: “Image registration techniques:
authors also found that a lot of research scope is present in a comprehensive survey”, (2014)
non-rigid image registration as it is closely related to the [19] Lijia Zh & Shaomin Zhang, “Research of non-rigid medical image
image fusion and super-resolution imaging and may help in registration based on gradient information”, International Symposium
building an image information system. on computational intelligence and design, (2017).
[20] Dung LR, Wu YY and Huang CM: “ Implementation of RANSAC
REFERENCES algorithm for feature-based image registration”. Journal of Computer
and Communications, Scientific Research Publishing 1(06), 46,
[1] Felix Ritter, Tobias Boskamp, Michael Schwier, Hendrik Laue: (2013).
“Medical image analysis”, IEEE Pulse, November, (2011).
[21] Sydoff, Marie, “Quantification methods for clinical studies in nuclear
[2] Abdolvahab Ehsani Rad, Alireza Norouzi, Ismail Bin Mat Amin, medicine - applications in AMS, PET/CT and SPECT/CT”, Medical
Mohd Shafry Mohd Rahim: “Principal techniques in medical image Radiation Physics Faculty of science, Lund University (2013)
analysis process” International Journal Of Interactive Digital Media,
[22] Shadrack Mambo, Yskandar Hamam, Barend van Wyk, Karim
1(3), 2289-4101, (2013).
Djouani and Patrick Siarry, “A review on medical image registration
[3] Sombir Singh Bisht, Bhumika Gupta, Parvez Rahi “Image registration techniques”, World Academy of Science, Engineering and
concept and techniques: a review”, Int. Journal of Engineering Technology International Journal of Computer and Information
Research and Applications, 4(7), 30-35, (2014) Engineering 12(1), (2018)
[4] Mohamed Abdel Basset, Arun Kumar Sangaiah, AhmedE. Fakhry, [23] Bernd Fischer and Jan Modersitzki. “Ill-posed medicine-an
Ibrahm El- henawy, Tie Qiu: “Feature and intensity based medical introduction to image registration”. Inverse Problems, 24(3), 03-40,
image registration using particle swarm optimization” J Med System, (2008)
41, 197, (2007).
[24] Kunio Doi, “Computer-aided diagnosis in medical imaging: historical
[5] Bankman, I. N., “Handbook of medical image processing and review, currentStatus and future potential”, Author manuscript;
analysis”. Academic Press, 2nd edition, (2008) available in PMC 2008 Jun 1
[6] Sebastian I. Sarvari, Kristina H. Haugaa, Wasim Zahid, Bjørn Bendz, [25] Alireza Norouzi, Ayman Altameem and Mohd Shafry Mohd
Svend Aakhus, Lars Aaberge, and Thor Edvardsen: “Layer-specific Rahim, ,“Medical image segmentation methods, algorithms and
quantification of myocardial deformation by strain echocardiography applications”, IETE TECHNICAL REVIEW | VOL 31 | NO 3 | MAY
may reveal significant CAD in patients with non–ST-segment JUN (2014)
elevation acute coronary syndrome”, JACC: Cardiovascular Imaging,
[26] Senthilkumaran, Thimmiaraja J, “Histogram equalization for image
6(5), 233-256, (2013)
enhancement using MRI brain images”, World Congress on
[7] Muthu Rama Krishnan Mookiah, Chua Kuang Chua Choo Min Lim Computing and Communication Technologies(2014)
E.Y.K.Ng, U. Rajendra Acharya ,Augustinus Laude: “Computer-
[27] I M Kazi, U. V. Kulkarni, “MRI brain image segmentation using
aided diagnosis of diabetic retinopathy: A review,” Computers in
adaptive thresholding and K-means algorithm”, International Journal
Biology and Medicine, 43(12), 2136-2155, (2013).
of Computer Applications (0975 – 8887) Volume 167 – No.8, June
[8] Kunio Doi: "Computer-Aided Diagnosis in medical imaging: (2017)
historical review, current status, and future potential" Comput. Med.
Imaging Graph. 31(4-5): 198–211, (2007).
[9] Nicholas Ayache, James S. Duncan: “Medical image analysis:
progress over two decades and the challenges ahead,” IEEE
Transactions on Pattern Analysis and Machine Intelligence. 22(1),
(2000).

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY KURUKSHETRA. Downloaded on August 07,2020 at 10:42:46 UTC from IEEE Xplore. Restrictions apply.