Computational Framework for Automated Disease Prediction Using Radiological Scans

Abstract

Automated disease prediction from radiological scan is one of the hot and promising research areas of AI for reliable and fast clinical diagnosis. In this paper, an efficient deep learning based automated computational framework is proposed for disease prediction using Computed Tomography, Magnetic Resonance Imaging and X-ray images. The proposed computational framework introduces various image processing techniques like resizing, normalization and data augmentation to enhance the quality of image and provides a standarized input for deep learning models training. Convolutional Neural Networkhas been utilized in the proposed computational framework to automatically detect several local features from radiological images and classify them into disease and healthy classes without handcrafted feature extraction step. The framework has been trained and tested on open source medical image benchmark datasets and the classification outcomes have been evaluated using several metrics such as accuracy, precision, recall, F1-score, etc. The results obtained from proposed system showed its reliable classification performance, while provides quick diagnosis, and gave a clinical decision support system. The proposed system not only infers a good computer-aided diagnosis system but also established a good platform for further research using transfer learning, explainable AI and multi-class disease prediction.