Machine Machine Learning in Medical Imaging In Machine learning

Machine Learning in Medical Imaging

Harikrishna Polavarapu (Mat. No. – 218060)

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

Medical Systems Engineering

Otto-von-Guericke University



Medical Imaging is a major field of medicine that makes doctor to understand the pathology and provide treatment for disease. The medical imaging modalities like X-ray, MRI, CT, and Ultrasound on daily basis generate large amount of data in clinics. The classification of this enormous data is not easy task. With recent developments, the application of machine learning algorithms has significantly increased in this field, mainly to categorize the data and to provide clinical support in diagnosis of various diseases. These algorithms with classical statistical and pattern recognition capabilities can build models by data that could help to predict and provide better diagnosis for many problems in medicine, with better accuracy than humans. In this paper overview of Support Vector Machine application in the field of medicine data classification is described.



Imaging is the processes of visualizing the information. Medicine imaging helps to study the nature of various diseases.  By using different imaging modalities for different purposes in medical examination .The digitalization of medical images produced by different techniques became easy with the development of Picture archiving and communication System 1. By Utilizing this data the computational and knowledge discovery capabilities of machine learning models and techniques could help providing better assistance for medical decisions. Machine intelligence based image interpretation has many applications in the field of medicine like predicting whether the produced image is normal or abnormal image 2.


Machine Learning in Medical Imaging

In Machine learning the main goal is to learn model from data. The algorithms are mainly two types namely supervised learning and unsupervised learning algorithms. In this report we mainly focus on supervised learning algorithms. In medicine supervised learning corresponds to classify the medical records into finite classes. Support Vector Machine is one of the supervised learning algorithms that are used for classification of data 3.

Inductive learning is the process behind the supervised learning algorithms. Initially the model is trained by sample set and evaluated. Given a new input to the model, it predicts whether the instance belong to particular target value or not. Machine learning in context of medical image classification there are many algorithms that can be applied for the same. When dealing with highly sensitive data like medical records the classification should be more accurate. D.B. in his approach evaluated the performance of SVM and fuzzy c-means clustering algorithm for classification of chest lesions. In this approach fuzzy c-means algorithm failed to reach average classification accuracy when compared to SVM with Gaussian RBF function 4.

Studies state that the SVM is having more accuracy and speed in classifying the data when compared to other super learning techniques such as Decision trees, neural networks, Naïve Bayes, K-NN. The Table below is the different Supervised Machine Learning algorithms that are used mainly for classification problems.

Table 1: Overview of different Machine learning algorithms that can used for Medical Image classification 3.


Support Vector Machine in Medical Imaging

Support Vector Machine (SVM) is famous in the field of bioinformatics that successfully performed DNA classification 5, protein classification 6 with high accuracy. With this the application of SVM became more dominant in biosciences.

In linear SVM the marginal boundaries that separate the two classes called Marginal Hyperplanes with Maximum margin between them are learned. By using these decision boundaries the Optimal Hyperplane is obtained. The Vectors are the supporting data points corresponding to the decision hyperplanes. These Vectors are therefore known as Support Vectors. Maximum margin between the decision surfaces results in the better generalization classification 7.

 In non-linearly separable case the low dimensional feature space is transformed to a larger dimensional space. But the computation cost in larger dimension Space is expensive in order to solve this problem Kernel Function like Gaussian Kernel, Polynomial kernel, RBF kernel, etc., are used. This is the basic explanation of Support Vector Machines in context of the papers we discuss below.



Implementation of SVM in the field of Medical Sciences

This section is consists of three effective implementation of Support vector machines in different approaches in medical imaging. The first application of SVM in medical data mining. The second application is about classification of CT images. The final application in classification of MRI images.


Application 1: Effectiveness of Support Vector Machine in Medical Data Mining 8

Data mining is the process of exploring data. In medicine context mining corresponds to the collection of information from the data. In this paper the application of Support Vector Machine in classifying data. And comparative study was performed with other machine learning classification algorithms like Naive Bayes and RBF Neural Network.

To evaluate the classification performance of various classifiers four different datasets were taken from UCI Machine learning Repository database. SPECT and STATLOG heart dataset, Breast datasets WISCONSIN and also PIMA diabetes dataset. For each dataset different attributes were taken into account.10-fold cross validation classification is applied (N=9).

Initially the models were trained and tested by different classifiers mentioned above. And the performance measures were calculated from the contingency table.

The performance of the SVM is calculated by accuracy, Specificity and sensitivity measures for WISCONSIN, SPECT, STATLOG and PIMA separately. If we consider the accuracy of SVM on above datasets are 93.75%, 90%, 84.44% and 90% respectively.

Finally from the performance measures we can conclude that the SVM performance is highly accurate when compared to other classification models like Naive Bayes and RBF networks.

Application 2: Quadratic program Optimization Using Support Vector Machine for CT brain image classifier. 9

This work proposed classification of CT brain images Using Support Vector Machine with Quadratic Optimization problem (QOP).QOP is the Optimization Constrain to maintain Maximum Margin between two decision surfaces.

As we already discussed that the SVM is a powerful classification model. In this SVM is used to categorize the CT brain MRI images to two different classes

For this task CT brain Images in total 150 images are considered as input data for SVM. Feature extraction is done from intensity profile of the given sample by finding Mean, Variance, Energy, Entropy, skewness, kurtosis.

The training and testing phase is performed on the model then the performance of the model is measured. The results showed that the performance measures of SVM like Accuracy, Sensitivity, and Specificity are 96.6%, 99% and 100% respectively.

The main drawback in this paper the sample size they consider is very low. For the same sample size other classifiers could also perform well. Finally, the results showed that the SVM is capable of classifying the CT MRI data with high performance measures.

Application 3: Brain MRI Slices Classification Using Least Squares Support Vector Machine. 10

In this paper a least squares support vector machine approach using Linear and RBF kernels to classify the Brain MRI slices into finite classes is explained. During the examination of brain MRI images the physician sometimes misjudges the tumor image as normal image when a large number MRI images were examined. This may lead to incomplete diagnosis of the patient. However results showed that use of machine learning technique in categorizing the slices is very helpful to solve this issue.

For this study they used a dataset containing 1100 MRI images out of which 833 are tumor images and 267 are normal images. Other Classification algorithms like K-NN, Multi-level Perceptron, and Radial Basis Function Neural Network are implemented for the same task and at the end the performance of different classifiers for the same data is compared.

Prior to the training and testing feature extraction is done by Gray level co-occurrence matric with the other statistical approaches like Mean, Variance, Entropy, Energy, Correlation calculations. The implementation is done in such a way the overall dataset is divided into two biased and unbiased datasets. For each datasets training and testing is performed. This procedure is repeated for all other classification techniques mentioned above.

By using confusion matrix the performance metrics like Sensitivity, Specificity and Accuracy are calculated respectively for each and every classification techniques. The results showed that for testing set 1 the LS-SVM with RBF kernel performed 99.9%, 95.5% and 98.64% respectively and LS-SVM with linear kernel performed 97.48%, 92.13% and 96.17% respectively. The LS-SVM performed outdatedly when compared to all other mentioned Classification techniques.

Similarly for testing set 2 Sensitivity, Specificity and Accuracy are calculated respectively for each and every classification techniques. The results showed that for testing set 2 the LS-SVM with RBF kernel performed 98.93%, 98.88% and 98.92% respectively and LS-SVM with linear kernel performed 98.16%, 96.66% and 97.98% respectively. The LS-SVM performed better compared to other classifiers.

Drawing conclusion from the results the LS-SVM is more advantageous to classify the medical data with high performance metrics.


In this report the overview of application of SVM in medical sciences along with three different published papers were presented. SVM performs outdatedly when compared to other classification models listed in this paper. The main difficulties in SVM are feature extraction because when large data sets are used features are very important to classify the data with high accuracy and also with large dataset the training takes more time for training the model. The other issues are with the increase in parameters (QPO) the computation complexity increases.


1Choplin R (1992). “Picture archiving and communication systems: an overview”.Radiographics. 12: 127-129.


2M. N. Wernick, Y. Yang, J. G. Brankov, G. Yourganov and S. C. Strother, “Machine Learning in Medical Imaging,” in IEEE Signal Processing Magazine, vol. 27, no. 4, pp. 25-38, July 2010.doi: 10.1109/MSP.2010.936730

3Supervised machine learning: A review of classification techniques-SB Kotsiantis – Informatica 31(2007) 249-268.

4 Hassen D.B., Taleb H., Yaacoub I.B., Mnif N. (2014) Classification of Chest Lesions with Using Fuzzy C-Means Algorithm and Support Vector Machines. In: Herrero Á. et al. (eds) International Joint Conference SOCO’13-CISIS’13-ICEUTE’13.

5Xiaoqiao Shen and Yaping Lin, “Gene expression data classification using SVM-KNN classifier ,”  Proceedings of 2004 International Symposium on Intelligent Multimedia, Video and Speech Processing, 2004., 2004, pp. 149-152.doi: 10.1109/ISIMP.2004.1434022

6″Support Vector Machines for predicting protein structural           class” Yu-Dong Cai, Xiao-Jun Liu, Xue-biao Xu and Guo-Ping Zhou-BMC Bioinformatics; 20012:3

7Ben-Hur A., Weston J. (2010),”A User’s Guide to Support Vector Machines. In Data Mining Techniques for the Life Sciences”. Methods in Molecular Biology (Methods and Protocols), vol 609. Humana Press doi:

8JANARDHANAN, Padmavathi; L., Heena; SABIKA, Fathima. Effectiveness of Support Vector Machines in Medical Data mining. Journal of Communications Software and Systems, S.l., v. 11, n. 1, p. 25-30, mar. 2015. ISSN 1846-6079.doi:

9J.Umamaheswari and Dr.G. Radhamani,Quadratic Program Optimization using Support Vector Machine for CT Brain Image Classification, International Journal of Computer Science Issues, Vol. 9, Issue 4, No 1, July 2012,

10 H. Selvaraj, S. Thamarai Selvi, D. Selvathi & L. Gewali Brain MRI Slices Classification Using Least Squares Support Vector Machine, International Journal of Intelligent Computing in Medical Sciences & Image Processing Vol. 1, Iss. 1,