U-Net-Attention-TBNet: A Cutting-Edge Solution for Accurate TB Lesion Segmentation and Classification

U-Net-Attention-TBNet is an innovative deep-learning model developed for precisely segmenting and classifying tuberculosis (TB) lesions in chest X-ray images. This Model combines the U-Net architecture with an advanced attention mechanism, enhancing feature extraction and boosting detection accuracy...

Full description

Saved in:
Bibliographic Details
Main Authors: Karthikeyan, M. Saravana, Raj, J. Relin Francis, Parvathi, R., Anushya, S. Thanga, Krishnan, R. Santhana, Joshua, K. Paul
Format: Conference Proceeding
Language:English
Subjects:
Accuracy
Attention Mechanism
Attention mechanisms
Biomedical imaging
Boosting
DenseNet
Feature extraction
Image segmentation
Lesions
Medical Image Classification
ResNet
RMSprop Optimizer
Robustness
Tuberculosis
U-Net
VGGNet
X-Ray Images
X-ray imaging
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:U-Net-Attention-TBNet is an innovative deep-learning model developed for precisely segmenting and classifying tuberculosis (TB) lesions in chest X-ray images. This Model combines the U-Net architecture with an advanced attention mechanism, enhancing feature extraction and boosting detection accuracy. Trained on the CheXpert dataset, U-Net-Attention-TBNet demonstrates superior performance compared to existing models, including DenseNet with U-Net, ResNet with U-Net, and VGGNet with U-Net. It achieves significantly higher accuracy, precision, recall, and F1 score, showcasing its effectiveness in distinguishing between primary TB lesions, secondary TB lesions, miliary TB lesions, and calcified granulomas. The attention mechanism refines the Model's ability to focus on pertinent features, leading to improved segmentation and reduced false positives. This progress represents a significant advancement in TB diagnosis, offering enhanced reliability and efficiency in medical imaging and contributing to better management of tuberculosis.
ISSN:2768-0673
DOI:10.1109/I-SMAC61858.2024.10714897