Classi cation of Thoracic Diseases Based on Chest X-ray Images using Kernel Support Vector Machine

Abstract

Machine learning is the leading eld of arti cial intelligence that has achieved expertlevel performance. Medical imaging has led to the advancements in diagnosis and treatment of various medical diseases. Classi cation or identi cation of thoracic diseases based on chest X-rays is one of the potential applications in medical imaging. The goal of this research work is the application of Support Vector Machine in the medical eld. The study consists of 112,120 images of 30,804 individual patients with fourteen thoracic disease labels. The SVM distinctively a ord balanced predictive performance and e ectively classify the diseases from image processing. To reduce the dimensionality and outliers from the SVM are coupled with novel Fast Principal Component Analysis (FPCA). SVM is accomplished in handling an in nite number of features and has a mathematical function named as kernels. Kernels address the non-linearity problem with accuracy and exibility. In this thesis, we assess the performance of four kernels of SVM namely Linear (L-SVM), Polynomial (P-SVM), Radial Basis (R-SVM), and Hyperbolic Tangent (H-SVM) for the classi cation of thoracic diseases based on X-ray images. Finally, based on the accuracy these four kernels are compared with each other and novel interpretations are also presented in the results part. It appears there is a signi cant (p ≤ 0.05) di erence between SVM kernels where P-SVM and R-SVM next in order outperform on most of the diseases identi cation models with average validated classi cation accuracy ranging from 90% to 98%. The average calibrated accuracy from 99.4% reaches 100% in most of the cases. Experimental results exhibit that SVM performs e ciently with kernels. The study is worth investigating as it is good for radiologists as they will be able to classify the diseases and it will help in improving and enhancing di erent medical techniques.