Optimised cascaded CNN for tuberculosis diagnosis: integrating bird swarm algorithm with advanced image segmentation

Abstract The rising global burden of tuberculosis (TB), specifically in resource-constrained areas, necessitates more reliable and efficient diagnostic solutions. Chest X-ray imaging, while commonly utilised for diagnosis, poses challenges owing to the complexity of TB systems and inconsistencies in the quality of image. This research introduces advanced image processing techniques with deep learning architecture to raise the precision and efficacy of TB prediction utilising chest X-ray. The initial preprocessing stage prepares the chest X-ray, involving resizing to a standard dimension followed by histogram equalisation to enhance image contrast. The improved bilateral filter (IBF) smooths the images with reduction in noise and preservation of edge information. The segmentation step employs an adaptive fuzzy c-means (AFCM) clustering technique for precise segmentation by isolating regions in chest X-ray. The segmented output features are extracted by grey-level co-occurrence matrix (GLCM) which assists in extracting important texture features. Finally classification is performed by bird swarm algorithm (BSA) optimised cascaded convolutional neural network (BSCasNet) for predicting TB. The BSA optimisation technique fine-tunes the cascaded CNN parameters utilising the collective and coordinated behaviour of birds flocking in search of food during migration for achieving more optimal solution. The execution of proposed TB detection system is analysed using the Python platform, and the results indicate that the proposed BSCasNet architecture results in an enhanced accuracy of 99.05% in classifying normal and tuberculosis compared to state-of-the-art approaches.