Tuberculosis detection using few shot learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Tuberculosis (TB) significantly impacts lung function, with Chest X-ray analysis being a primary detection method.
• Traditional Deep Learning models for TB detection are often too large and computationally intensive for practical deployment.
• Generating augmented datasets for medical imaging is not recommended due to potential inaccuracies.

Purpose of the Study:

• To develop an efficient and accurate model for Tuberculosis detection using Chest X-ray analysis.
• To address the limitations of traditional Deep Learning models in terms of size, parameters, and computational requirements.
• To enable real-time TB detection on edge devices for improved accessibility and efficiency.

Main Methods:

• A novel model, TB-FSNet, was developed, integrating Few Shot Learning - Prototypical Network (FSL-PT) with a modified MobileNet-V2 backbone and a Self-Attention layer.
• TB-FSNet was trained using the FSL-PT paradigm with six different backbones.
• The study utilized a combined dataset from the Montgomery County and Shenzhen Chest X-ray datasets.

Main Results:

• The proposed TB-FSNet model achieved a highest accuracy of 93.6%.
• The model boasts a small footprint with only 2.21M parameters and 8.67 MB size.
• High performance metrics including precision, specificity, and sensitivity were maintained.

Conclusions:

• TB-FSNet offers an efficient and accurate solution for Tuberculosis detection from Chest X-rays.
• The model's lightweight design makes it suitable for deployment on embedded and edge devices for real-time analysis.
• This advancement enhances the accessibility and efficiency of TB detection, providing immediate diagnostic insights.