RNGU-NET: a novel efficient approach in Segmenting Tuberculosis using chest X-Ray images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Tuberculosis (TB) is a significant global health issue, affecting vital organs like the lungs, kidneys, and brain.
• The World Health Organization (WHO) reported approximately ten million TB infections in 2020.
• Accurate segmentation of TB in medical images is crucial for diagnosis and treatment.

Purpose of the Study:

• To introduce and evaluate a novel deep learning architecture, RNGU-NET, for improved tuberculosis segmentation.
• To address the limitations of existing U-Net based models in medical image segmentation.
• To compare the performance of RNGU-NET against U-NET and U-NET+ResNet.

Main Methods:

• Development of the RNGU-NET architecture, integrating ResNet, Non-Local Block, and Gate Attention Block.
• Enhancement of the encoder with ResNet and the decoder with Gate Attention Block.
• Introduction of a Local Non-Local Block to mitigate the bottleneck issue in U-Net models.
• Comparative analysis using the Shenzhen dataset for tuberculosis segmentation.

Main Results:

• RNGU-NET achieved superior performance in tuberculosis segmentation.
• The proposed RNGU-NET model demonstrated an accuracy rate of 98.56%, a Dice coefficient of 97.21%, and a Jaccard index of 96.87%.
• U-NET and U-NET+ResNet showed lower performance metrics compared to RNGU-NET.

Conclusions:

• The RNGU-NET architecture represents a significant advancement in tuberculosis segmentation.
• The novel architectural components effectively overcome previous segmentation model limitations.
• RNGU-NET shows high potential for clinical application in diagnosing and monitoring tuberculosis.