ACL

ACL segments ground-glass opacities (GGOs) in lung computed tomography (CT) images to improve delineation of low-contrast lesions such as those associated with COVID-19.

Key Features:

• Attention Mechanism Threshold: Employs an attention mechanism to dynamically adjust threshold parameters based on image contrast and to divide the lung into three contrast-defined regions.
• Contour Equalization: Applies contour equalization techniques to enhance edge definition and facilitate more accurate delineation of GGO boundaries.
• Lung Segmentation: Performs lung segmentation to confine analysis to the segmented lung area and exclude irrelevant structures.
• Adaptive Threshold Adjustment: Fine-tunes segmentation thresholds for different lung regions according to local contrast characteristics.
• Performance Improvements: Reported results include an 8.9% improvement in Dice similarity coefficient and a 23% reduction in average symmetry surface distance (ASD) on four COVID-19 datasets.
• Computational Efficiency: Demonstrates markedly lower computational cost relative to deep learning models (reported as 0.09% of their computational power).

Scientific Applications:

• COVID-19 diagnostics and monitoring: Enables precise segmentation of GGOs for assessment of COVID-19 lung involvement.
• Disease progression and treatment response assessment: Supports quantitative evaluation of changes in GGO extent and boundaries over time.
• Analysis of low-contrast CT datasets: Facilitates evaluation of diverse patient datasets where GGOs present with low intensity contrast.

Methodology:

Adaptive threshold adjustment tailored to image contrast; attention mechanism thresholds fine-tuned per lung region with a three-region division; contour equalization for edge enhancement; and lung segmentation to restrict processing to the segmented lung area.