Application of medical informatics and data analysis methods for automatic medical diagnosis

Abstract

With the increased size and complexity of data, interest has rapidly emerged in adopting artificial intelligence (AI) and deep learning (DL) to create data-driven models for automating medical diagnosis, and for neuromuscular disorders (NMDs) in particular. Therefore, this thesis aims to address the gaps in this context. To provide clinicians with a more objective and accurate methods for assessing muscle fatigue, a convolutional neural network (CNN)-based DL model was proposed to classify simulated surface electromyography (EMG) signals into different maximum voluntary contraction levels, achieving and accuracy of 88.88%. To ensure transparency and clinicians trust, the interpretability of Multi-Layer Perceptron (MLP) and Residual Neural Network (ResNet)-based DL models that achieved 95.67% and 98.37% testing accuracies, respectively, for myopathy diagnosis, was investigated. Shapley additive explanation (SHAP) for feature-based interpretation, and Gradient-weighted class activation mapping (Grad-CAM) for visual interpretation of raw signals, were employed, providing clear insights into the decision-making process. Furthermore, given the recent emergence and proved ability of quantum machine learning to handle high-dimensional data and solve complex tasks, a study was introduced to explore its potential in myopathy diagnosis. Quantum support vector machines (QSVMs) with variational quantum circuit-based kernels were proposed, and their performance was compared with classical methods. A hybrid QSVM model trained on deep features demonstrated promising classification ability, with training and testing accuracies of 96.7% and 85.1%, respectively. The results obtained in our research shed new light on the application of medical informatics in the field of healthcare and the EMG-based NMDs diagnosis in particular, indicating promising potential for future adoption of automated medical decision-making