Detection of Tuberculosis from Chest X-ray using Deep Transfer Learning
Abin Antony, Khaled Tawfiq Al-Assaf, Thangamuthu Kumaran Vaiyapuri, K. Vijayakumar, D Keerthivasan, Alok Jain, Md. Tabil Ahammed
Abstract
Tuberculosis (TB) is a severe lung infection and needs early recognition and treatment. Chest X-ray is a common imaging technique for screening TB and the outcome helps to detect the severity of the tb in a person. Manual examination is quite difficult and hence, most of the recent works considered the automatic detection of the TB using deep-learning (DL) models. This work proposed a DL-tool with fused-features to classify the X-ray into healthy/TB with softMax classifier. The stages of this approach include; image collection and initial adjustment, deep-fetaures extraction with DL-models and SoftMax classification, best model features identification and generating fused-features (FF) using 50% dropout and serial features fusion, and classification with the developed system. This system offered 98% accuracy when FF-based detection is performed, which confirms the merit of this system.