AI based Disease Diagnosis using Medical Imaging

Abstract

The question is not whether intelligent machines can have any emotions, newline but whether machines can be intelligent without any emotions.- Marvin Minsky, 1986 newline This work presents various methods of automated analysis of medical images to detect the newline abnormalities associated with a disease. Knee X-rays, chest X-rays, and brain MR images newline are the chosen areas of focus. The proper and accurate study of medical images require newline human expertise and time. This study aims to build an aid that helps in decision making by newline providing a quick automated diagnosis. In this work, AI based methods for fully automated newline disease diagnosis have been proposed. Knee osteoarthritis is a common degenerative disease newline among elderly people all over the world. Automated analysis of AP (anteroposterior) and newline lateral views of knee X-ray images is used to detect this disease. For detection of patellar newline osteophytes, patella region of knee X-ray is segmented using entropy based segmentation newline method. The features extracted by convex hull based and chain code based approaches newline are fed to a pre-trained binary support vector machine to identify patellar osteoarthritis. newline Detection of low grade osteoarthritis (grade 1 and 2) from X-ray images is a challenging newline problem. A lightweight convolutional neural network has been proposed here to detect lower newline grade osteoarthritis. A small region-of-interest is detected and analyzed by a dual-path dual newlinekernel based module that utilizes depth-wise separable convolution. Using a very limited newline system resources, the proposed architecture gives satisfactory result in classifying the lower newline grade osteoarthritis. Lung diseases are now considered a global threat in the post-COVID newline age. Analysis of chest X-ray is essential in detection of lung infections. A hybrid classifier is newline proposed to diagnose four different types of lung infections. In the proposed model, transfer newline learning with fine-tuning on DenseNet169 and MobileNet architecture is used for feature newline extraction. Feature dimension is optimized by utilizing the conc