Enhanced CBAM-Efficient Net Model for Efficient Tuberculosis Diagnosis Using Chest X-Ray Images
Abstract
The CBAM-Efficient Net Model integrates the Convolutional Block Attention Module (CBAM) with the Efficient Net architecture for better focus on relevant regions of the images for precise detection of tuberculosis (TB) from chest X-rays. Built from scratch with X-rays from Kaggle, it utilizes data augmentation (image compression, elastic transformation), contrastive learning, and advanced feature extraction to enhance performance. In the final stage, Vision Transformers in a hybrid architecture improves the model's accuracy. In addition to significance visualization, Grad-CAM offers clinicians an attention visualization. Post-training quantization and pruning help keep the model compact and efficient for use in clinical settings. The system is designed to perform TB diagnosis predictions in real-time through a Flask interface with ngrok. Keywords: TB detection, Deep Learning, CBAM, Efficient Net, Vision Transformer, Grad-CAM, Chest X-Ray
Cite this paper
Published in
