Class dependency based learning using Bi-LSTM coupled with the transfer learning of VGG16 for the diagnosis of Tuberculosis from chest x-rays

TL;DR

This paper introduces an automated TB diagnosis method from chest X-rays using lung segmentation, transfer learning with VGG16, and Bi-LSTM, achieving high accuracy and sensitivity on public datasets.

Contribution

It combines lung segmentation with a novel Bi-LSTM and VGG16 transfer learning approach for improved TB detection from X-rays.

Findings

Achieved 97.76% accuracy on Schezien dataset.

Enhanced diagnostic accuracy by 0.7% and 11.68% on two datasets.

Demonstrated effective lung segmentation and feature extraction for TB diagnosis.

Abstract

Tuberculosis is an infectious disease that is leading to the death of millions of people across the world. The mortality rate of this disease is high in patients suffering from immuno-compromised disorders. The early diagnosis of this disease can save lives and can avoid further complications. But the diagnosis of TB is a very complex task. The standard diagnostic tests still rely on traditional procedures developed in the last century. These procedures are slow and expensive. So this paper presents an automatic approach for the diagnosis of TB from posteroanterior chest x-rays. This is a two-step approach, where in the first step the lung regions are segmented from the chest x-rays using the graph cut method, and then in the second step the transfer learning of VGG16 combined with Bi-directional LSTM is used for extracting high-level discriminative features from the segmented lung regions and then classification is performed using a fully connected layer. The proposed model is evaluated using data from two publicly available databases namely Montgomery Country set and Schezien set. The proposed model achieved accuracy and sensitivity of 97.76%, 97.01% and 96.42%, 94.11% on Schezien and Montgomery county datasets. This model enhanced the diagnostic accuracy of TB by 0.7% and 11.68% on Schezien and Montgomery county datasets.