Texture Feature Selection Using GA for Classification of Human Brain MRI Scans
Intelligent Medical Image Analysis plays a vital role in identification of various pathological conditions. Magnetic Resonance Imaging (MRI) is a useful imaging technique that is widely used by physicians to investigate different pathologies. Increase in
- PDF / 1,229,730 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 11 Downloads / 214 Views
Department of Electrical Engineering, Comsats Institute of Information Technology (CIIT), Abbottabad, Pakistan [email protected], [email protected], [email protected] 2 Department of Electronics Engineering, International Islamic University (IIUI), Islamabad, Pakistan [email protected]
Abstract. Intelligent Medical Image Analysis plays a vital role in identification of various pathological conditions. Magnetic Resonance Imaging (MRI) is a useful imaging technique that is widely used by physicians to investigate different pathologies. Increase in computing power has introduced Computer Aided Diagnosis (CAD) which can effectively work in an automated environment. Diagnosis or classification accuracy of such a CAD system is associated with the selection of features. This paper proposes an enhanced brain MRI classifier targeting two main objectives, the first is to achieve maximum classification accuracy and secondly to minimize the number of features for classification. Two different machine learning algorithms are enhanced with a feature selection pre-processing step. Feature selection is performed using Genetic Algorithm (GA) while classifiers used are Support Vector Machine (SVM) and K-Nearest Neighbor (KNN). Keywords: Feature selection Brain MRI Genetic algorithm Support vector machine Classifier Machine learning Supervised learning
1 Introduction Since MRI does not use any ionizing radiation, it is generally safe compared to CT (Computed Tomography) scan. Manual diagnosis of medical images is subjective, time consuming and costly. Looking at an image (visual perception) and interpreting what is seen is often prone to errors due to technician oversights [1]. At the same time due to various imaging constraints and tissue characteristics, automated classification of brain MRI into normal and abnormal studies is also quite difficult. Classification is an automated process that intends to order every information/ data/instance in specific class, in light of the data portrayed by its features. However, without previous knowledge, useful features cannot be determined for classification. So initially it requires an introduction of large number of features for classification of a particular dataset. Introducing a large number of features may include irrelevant and © Springer International Publishing Switzerland 2016 Y. Tan et al. (Eds.): ICSI 2016, Part II, LNCS 9713, pp. 233–244, 2016. DOI: 10.1007/978-3-319-41009-8_25
234
M.N. Tajik et al.
redundant features which are not helpful for classification and this can even lessen the performance of a classifier due to large search space known as “the curse of dimensionality” [2]. This problem can be subsided by selecting just relevant features for grouping. By omitting irrelevant and unnecessary features, feature selection reduces the training time and minimizing the feature set, thus improving the performance of classifier [3, 4]. During the analysis of tissue in MRI by radiologists, image texture plays a pre-dominant role. In fact texture (in) homogeneity is
Data Loading...
