SRIS: Saliency-Based Region Detection and Image Segmentation of COVID-19 Infected Cases.

Area of Science:

• Computer Vision
• Medical Image Analysis
• Image Processing

Background:

• Image noise and artifacts degrade segmentation model performance.
• Intensity inhomogeneity presents challenges for accurate image segmentation.

Purpose of the Study:

• To propose a novel saliency-based region detection and image segmentation (SRIS) model.
• To enhance image segmentation robustness against noise and intensity variations.
• To improve contour initialization independence in level-set evolution.

Main Methods:

• Developed a novel adaptive level-set evolution protocol.
• Designed an adaptive weight function within the level-set energy function.
• Implemented sign modulation of the energy function to mitigate noise effects.

Main Results:

• The SRIS model demonstrated superior performance on complex real and synthetic images compared to existing models.
• Statistical analysis on COVID-19 CT images and THUS10000 dataset confirmed high segmentation accuracy and time efficiency.
• The model proved robust to contour initialization variations.

Conclusions:

• The proposed SRIS model effectively addresses challenges in image segmentation caused by noise and intensity inhomogeneity.
• SRIS offers a promising tool for accurate and efficient medical image analysis, including early COVID-19 screening.
• The model's robustness and efficiency make it suitable for practical applications.