E-TBNet: Light Deep Neural Network for Automatic Detection of Tuberculosis with X-ray DR Imaging.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Computer-aided diagnosis

Background:

• Tuberculosis (TB) detection using chest X-rays faces challenges with hardware demands and limited labeled data.
• Existing deep neural network models struggle with training due to data scarcity and specific image characteristics.

Purpose of the Study:

• To develop an efficient and accurate tuberculosis detection model deployable on low-cost personal computers and embedded devices.
• To address the limitations of current TB detection models regarding hardware requirements and data availability.

Main Methods:

• Optimized a residual module and incorporated an attention mechanism for channel feature fusion.
• Designed and adjusted a novel network architecture, E-TBNet, for efficient TB screening.
• Analyzed data distribution of public TB datasets to inform model development.

Main Results:

• E-TBNet demonstrated superior recall and accuracy compared to SqueezeNet and ShuffleNet on PC and Jetson Xavier embedded devices.
• The proposed model achieved a shorter reasoning time, indicating higher efficiency.
• Experimental results confirmed the effectiveness of the optimized network architecture.

Conclusions:

• E-TBNet offers an efficient and accurate solution for tuberculosis screening on hardware-limited devices.
• The model's performance and efficiency make it suitable for widespread deployment in resource-constrained environments.
• The study highlights the potential of optimized deep learning models for accessible medical diagnostics.