A Lightweight Hybrid Deep Learning Model for Tuberculosis Detection from Chest X-Rays.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Tuberculosis (TB) poses a significant global health challenge, especially in resource-limited areas.
• Early and accurate detection of TB using chest X-ray imaging is crucial for reducing mortality and transmission.
• Existing diagnostic methods may lack the speed or accuracy required for widespread screening.

Purpose of the Study:

• To introduce a novel hybrid deep learning approach for enhanced tuberculosis detection from chest X-ray images.
• To develop a model that balances high classification accuracy with computational efficiency for practical deployment.
• To leverage the strengths of both convolutional neural networks and transformer models for robust feature extraction.

Main Methods:

• A hybrid deep learning architecture combining GhostNet (efficient CNN) and MobileViT (transformer) was developed.
• Feature fusion techniques were employed, integrating spatial features from GhostNet and contextual representations from MobileViT.
• The model was trained and evaluated on two public chest X-ray datasets, comparing its performance against state-of-the-art CNN architectures.

Main Results:

• The hybrid model achieved high accuracy, reaching 99.52% on dataset 1 and 99.17% on dataset 2.
• The model demonstrated a low computational cost, with 7.73M parameters and 282.11M Floating Point Operations.
• Performance surpassed individual baseline models, indicating the effectiveness of the hybrid approach.

Conclusions:

• Feature-level fusion between CNN and transformer branches enables robust tuberculosis detection.
• The proposed model offers high accuracy with low inference overhead, suitable for clinical settings.
• The lightweight design makes this model ideal for deployment in resource-constrained environments.