Optimized Detection of Tuberculosis Through Contour-Aware Superpixel Segmentation and ConvBiLSTM with Knowledge Distillation.

The critical challenge of tuberculosis (TB) detection by introducing an advanced
methodology combining contour-aware superpixel segmentation and Convolutional
Bidirectional Long Short-Term Memory (ConvBiLSTM) networks, enhanced with knowledge
distillation. The approach involves precise segmentation of lunge X-ray images to isolate relevant
features, followed by processing with a ConvBiLSTM model to capture both spatial and temporal
information. Knowledge distillation further improves model efficiency by training a smaller
model to mimic a more complex one. Evaluated on multiple datasets, the proposed method
demonstrates significant improvements in TB detection accuracy and computational efficiency,
highlighting its potential for early and reliable TB diagnosis.
Keywords: ConvBi-LSTM, Knowledge Distillation, Tuberculosis Classification, chest X-ray
images.