Automatic Healthcare Diagnosis and Prediction Assessment based on AI Multi-Classification Algorithm

This work focuses on the concept of Medical Decision Support System (MDSS) to support doctors/physicians and act as a secondary validation tool in the process of diagnosis and disease prediction. In this paper, we conduct a comparative study on a multi-class classification algorithm that utilizes X-ray and CT-scan imagery inputs for the automatic diagnosis of COVID-19, and pneumonia against healthy cases. We consider the flexible implementation of a hybrid algorithm that combines two versions of pre-trained Neural Networks with five different classification learners, including Support Vector Machine (SVM), K-Nearest Neighbor (KNN), discriminant, tree, and Naive Bayes. In addition, we also experimented with six different coding designs, including ternary complete, all pairs, binary complete, one vs. all, one vs. one, and ordinal. The goal of this study is to assess the performance of different learners and coding designs to identify the best version of the hybrid algorithm. We found that the pre-trained network Resnet-50 with SVM classification learner and ternary complete delivered the best performance accuracy rate of 98.7%.