Multiclass Classification of Disease Using CNN and SVM of Medical Imaging

Abstract

AbstractThis paper proposed a model that deals with automatic prediction of the disease given the medical imaging. While most of the existing models deals with predicting disease in one part of the body either brain, heart or lungs, this paper focuses on three different organs brain, chest, and knee for better understanding the real word challenge where problems do not include crisp classification but the multiclass classification. For simplicity this paper focuses on just determining whether that organ is affected with the disease or not and future work can be done by further expanding the model for multiple disease detection of that organ. We have used CNN for multiclass image classification to determine the input medical image is brain, chest or knee and then SVM is used for binary classification to determine whether that input image is detected with the disease or not. Three different datasets from Kaggle are used: Brain Tumor MRI Dataset, COVID-19 Chest X-ray Image Dataset and Knee Osteoarthritis Dataset with KL Grading. Images from these datasets are used to make fourth datasets for training and testing the CNN for the prediction of the three different organs and after that output will be the input of respective SVM classifier based on the output result and predict the weather it is diagnostic with the disease or not. The proposed model can be employed as an effective and efficient method to detect different human diseases associated with different parts of the body without explicitly giving the input that it belongs to that part. For the transparency this model displays the accuracy of prediction made for the input image.KeywordsCoronavirusBrain tumorOsteoarthritisCNNSVMMedical images