Microscopic Image-Based TB Detection Using an Enhanced Bi-LSTM Model Optimized by Firefly Algorithm

Abstract-Tuberculosis (TB) remains a major global health concern, particularly in resource-limited regions where
early and accurate diagnosis is crucial. This study proposes a novel deep learning-based approach for TB detection
using microscopic sputum smear images. The core of the system is an enhanced Bidirectional Long Short-Term
Memory (Bi-LSTM) model, tailored to capture complex sequential patterns within medical image features. To
improve classification accuracy and convergence efficiency, the Firefly Algorithm is employed to optimize the
model’s hyperparameters. Preprocessing techniques, including noise reduction and contrast enhancement, are applied
to improve image quality, followed by feature extraction using convolutional layers. The optimized Bi-LSTM model
is trained and validated on a curated dataset of TB-positive and TB-negative images. Experimental results demonstrate
superior performance in terms of accuracy, sensitivity, and specificity compared to conventional models. This
framework highlights the potential of integrating evolutionary optimization with deep learning for robust TB diagnosis
in clinical settings.
Keywords: Tuberculosis detection, Bi-LSTM, Firefly Algorithm, microscopic images, deep learning, medical
image analysis, optimization, sputum smear.