Date of Award
1-1-2025
Document Type
Thesis
Degree Name
M.S. in Engineering Science
First Advisor
Kasem Khalil
Second Advisor
Md Sakib Hasan
Third Advisor
Elliott Hutchcraft
School
University of Mississippi
Relational Format
dissertation/thesis
Abstract
The growing demand for accurate, rapid, and interpretable diagnostic tools in the medical field has spurred the integration of artificial intelligence into various healthcare applications. Acute Myocardial Infarction, which remains one of the primary causes of mortality in both developed and developing nations, requires timely and precise identification to prevent life-threatening complications. In response to the lack of interpretable AI models tailored to diagnosing AMI using hemogram parameters, this study presents a robust machine learning-based framework that explores the diagnostic potential of these parameters. By evaluating several machine learning techniques, the research establishes the superiority of boosting algorithms, particularly the Light Gradient Boosting Machine, in identifying AMI cases. The proposed model achieves a commendable accuracy of 85.31% and an Area Under the Curve score of 85.28%, signifying its high discriminative power between AMI patients and healthy individuals, while simultaneously offering transparency in decision-making—an essential factor in clinical applications.
Extending the application of AI in cardiovascular research, this study also addresses the critical challenge of predicting multiple complications associated with heart disease, such as cardiogenic shock, pulmonary edema, and ventricular fibrillation, among others. A novel framework combining a K-Nearest Neighbors-based imputation technique with a Gradient Boosting model is proposed for multi-label classification of heart disease complications. The methodology demonstrates significant improvement over conventional techniques by employing an enriched pre-processing pipeline and a rigorous comparative analysis across several machine learning algorithms. Evaluation through metrics including accuracy, precision, recall, F1-score, and Hamming loss affirms the effectiveness and generalizability of the framework. Moreover, a novel framework is proposed that leverages a Hybrid Shallow Neural Network (HSNN) architecture, designed to balance model complexity and computational efficiency. Additionally, a new feature selection algorithm, termed Gini Importance for Multi-Label (GIML), is introduced, which systematically evaluates and selects the most relevant features across all labels by employing a gini impurity-based mechanism. Furthermore, in the domain of biomedical imaging, particularly cell nucleus segmentation in microscopy images, the study introduces mA-UNet—an advanced model engineered to detect fine-grained foreground elements in imbalanced datasets. This model outperforms existing approaches, attaining a mean Intersection over Union score of 95.50%, and exhibits hardware efficiency when deployed on the Zynq UltraScale+ FPGA. Collectively, the research underscores the transformative role of interpretable and high-performance AI in addressing complex clinical and biomedical imaging tasks.
Recommended Citation
Kader Khan, Md Rahat, "An Advanced Machine Learning Framework and Multi-Label Heart Disease Classification with UNet-Based Nucleus Segmentation" (2025). Electronic Theses and Dissertations. 3306.
https://egrove.olemiss.edu/etd/3306