A Novel Approach for the Detection of Tuberculosis and Pneumonia Using Chest X-Ray Images for Smart Healthcare Applications

The early discrimination of pneumonia and tuber-culosis is difficult for radiologists due to similar pathological symptoms in the chest X-ray images. Therefore, there is a requirement for automated methods to detect and classify these two diseases accurately. This letter proposes the multiscale eigendomain gradient boosting (MEGB)-based approach to detect pneumonia and tuberculosis from chest X-ray images. The discrete wavelet transform is employed to evaluate subbands of chest X-ray images at different decomposition levels. The singular value decomposition (SVD) is utilized in each subband to evaluate singular values, left eigenmatrix, and right eigenmatrix, respectively. The maximum value of each column of both left and right eigenmatrices and singular values for each subband of the X-ray image are used as features. All subband eigendomain feature vectors are concatenated and given to the light gradient boosting model to detect pneumonia and tuberculosis diseases. The performance of the proposed MEGB approach-based detection is evaluated using chest X-ray images from a publicly available database. The suggested MEGB approach has achieved an accuracy value of 96.42%. The suggested approach performs better than the transfer learning and other reported methods to detect pneumonia and tuberculosis using chest X-ray images.