Deep learning models achieved state-of-the-art performance in many fields including biomedical due to the ability of convolutional network (CNN) models and knowledge of transfer learning approaches. The CNN models gave scope for tuberculosis classification by enabling transfer learning approaches. In this paper, we analyzed the effect of transfer learning techniques on pre-trained deep learning models, which have less feature maps when compared to existing models for classification of chest X-ray images of potential tuberculosis patients. The layer-by-layer performance was observed, and the effect of transfer learning was analyzed in terms of different tuning modes: (a) shallow tuning, (b) fine-tuning and (c) deep tuning, where deep tuning showed the best performance in terms of sensitivity, specificity, ROC-AUC and accuracy.
G. Babu, K, S. Saj T., Vishvanathan, S., and Dr. Soman K. P., “Tuberculosis Classification Using Pre-trained Deep Learning Models”, in Advances in Automation, Signal Processing, Instrumentation, and Control, Select Proceedings of i-CASIC 2020, 2021, pp. 767-774.