Pathological Asymmetry-Guided Progressive Learning for Acute Ischemic Stroke Infarct Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate quantitative infarct estimation is vital for acute ischemic stroke (AIS) diagnosis, treatment, and prognosis.
• Early ischemic changes are subtle and easily confused with normal brain tissue, making segmentation difficult.
• Existing methods often fail to adequately utilize anatomical asymmetry and domain knowledge, leading to mis-segmentation.

Purpose of the Study:

• To develop an advanced method for accurate AIS infarct segmentation.
• To address the limitations of current methods in handling anatomical asymmetry and domain knowledge.
• To propose a novel pathological asymmetry-guided progressive learning (PAPL) method.

Main Methods:

• PAPL employs a three-stage progressive learning approach: knowledge preparation, formal learning, and examination improvement.
• The knowledge preparation stage uses contrastive learning to enhance pathological asymmetry discrimination.
• The formal learning stage incorporates a feature compensation module (FCM) for anatomical context aggregation, and the examination improvement stage uses a perception refinement strategy (RPRS) for correction.

Main Results:

• The PAPL method demonstrated superior performance in AIS infarct segmentation.
• Experiments on public and in-house NCCT datasets validated the effectiveness of the proposed approach.
• The method successfully improved segmentation accuracy by leveraging pathological asymmetry and progressive learning stages.

Conclusions:

• The proposed PAPL method offers a promising solution for accurate AIS infarct segmentation.
• This approach has the potential to significantly aid in stroke evaluation and treatment planning.
• PAPL's ability to learn domain-specific knowledge and refine predictions marks a significant advancement in stroke imaging analysis.