In recent decades, the integration of Computer-Aided Diagnosis (CAD) systems is crucial in such contexts, offering a cost-effective alternative to expert medical assessment and addressing disease detection concerns in resource-constrained settings. This thesis presents an innovative machine learning model for multi-class chest X-ray diagnosis. Unlike previous binary classification studies, our work uniquely addresses the classification of 15 distinct classes, demonstrating a broader and more complex application in disease detection.
Utilizing the NIH chest X-ray dataset, our approach involves comprehensive data preprocessing, including image resizing, augmentation, and segmentation. The core of our model combines deep feature extraction via ResNet architecture and feature selection enhanced by AAPSO. Crucially, our model integrates an augmented EfficientNetB0 architecture, extended with 12 additional layers for superior performance.
Our
deep CNN model leverages transfer learning and achieves a notable 87.9%
accuracy, significantly enhancing chest X-ray disease classification
efficiency. This marks a considerable leap over standard chest X-ray disease
classification approaches, showcasing our model's potential in transforming
chest X-ray diagnostics in resource-limited settings