Ensemble Technique Coupled with Deep Transfer Learning Framework for Automatic Detection of Tuberculosis from Chest...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Public Health

Background:

• Tuberculosis (TB) is a leading global cause of death, necessitating efficient screening methods.
• Chest X-rays (CXRs) are crucial for TB diagnosis, but manual interpretation can be challenging.
• Automated analysis of CXRs using deep learning (DL) offers a promising approach for early TB detection.

Purpose of the Study:

• To develop and evaluate an automated deep learning system for tuberculosis detection from chest X-rays.
• To investigate the impact of image segmentation on the performance of DL models for TB classification.
• To compare the efficacy of various convolutional neural network (CNN) models and an ensemble approach for TB diagnosis.

Main Methods:

• Utilized the U-Net model for segmenting relevant lung regions in CXRs, followed by feature extraction.
• Employed eight different CNN models for classification tasks on segmented and un-segmented CXR datasets.
• Developed and tested a stacked ensemble algorithm integrating multiple DL models for enhanced diagnostic performance.

Main Results:

• The U-Net model achieved high segmentation accuracy (98.58%), IoU (93.10), and Dice score (96.50).
• The proposed stacked ensemble model demonstrated superior performance with 98.38% accuracy, 98.89% sensitivity, and 98.70% specificity.
• Segmented CXR images processed with ensemble learning significantly outperformed un-segmented images in TB detection.

Conclusions:

• Automated TB detection using DL on segmented CXRs is a viable and effective screening tool.
• Image segmentation using U-Net enhances the performance of DL models in identifying TB from chest X-rays.
• Ensemble learning approaches combined with image segmentation offer a robust solution for improving TB diagnostic accuracy.