Tuberculosis Detection in X-Ray with Two-Fold Training and Fused Deep Features Based Classification

Tuberculosis (TB) is a communicable lung infection and needs opportune recognition and treatment. Clinical level screening of the TB is frequently executed using the chest X-rays (CXR) and its examination help to identify the severity of the disease. Recently, computer algorithm supported diagnosis of the CXR is common to automate the disease diagnosis and treatment planning process. This study aims to employ a Deep-Learning (DL) tool to detect the TB from CXR dataset. To improve the accuracy, a Two-Fold Training (TFT) process is considered. Various stages in this DL-tool includes; (i) CXR collection, resizing, and enhancement, (ii) feature extraction using DL-model using TFT, (iii) implementing classification with SoftMax and identification of best two DL-models based on the detection accuracy, (iv) 50% dropout based feature reduction, serial features integration to generate Fused-feature Vector (FV), and (v) classification and performance confirmation using 3-fold cross validation. The results of this study substantiate that, the PDL-scheme helped to attain better accuracy (100%) when FV and random-forest based classification is executed.