Serially Fused Dual-Deep-Features Based Chest X-Ray Classification Scheme to Detect Tuberculosis

In hospitals, the increased frequency of disease occurrence has resulted in an increased diagnostic burden; therefore, several semi/fully automatic disease detection systems have been developed in order to decrease this burden. Through Serially Fused Dual-Deep-Features (SFDDF), we propose a framework for classifying chest X-rays into healthy or TB classes based on a Tuberculosis (TB) detection framework. This scheme consists of several phases, including the collection, modification, and enhancement of images, the extraction of deep features (DF) with pre-trained schemes, optimization of features with the bat algorithm, and the generation and validation of SFDDFs. As part of the evaluation process for this paper, 3000 benchmark X-ray images (1500 healthy and 1500 TB) were considered, and classification procedures were performed using (i) Individual DFs and (ii) SFDDFs, and the results were verified. Based on the X-ray image database, the proposed SFDDF with random forest (RF) classifier has an accuracy rate of >98%, which confirms its significance in the detection of TB based on X-rays.