Semi-Supervised Medical Hyperspectral Image Segmentation Using Adversarial Consistency Constraint Learning and Cross...

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Hyperspectral imaging offers rich spatial and spectral data for pathological image segmentation.
• Manual annotation of medical hyperspectral images is labor-intensive and time-consuming, hindering model training.
• There is a critical need for semi-supervised learning frameworks to leverage unlabeled data.

Purpose of the Study:

• To develop a novel semi-supervised learning framework for accurate pathological image segmentation using hyperspectral data.
• To address the challenge of limited annotated data in medical hyperspectral image analysis.

Main Methods:

• Proposed the adversarial consistency constraint learning cross indication network (ACCL-CINet).
• Employed a spatial-spectral feature encoding approach with contextual and structural encoders.
• Utilized a cross indication attention module for feature aggregation.
• Implemented a semi-supervised training strategy with pixel perceptual consistency and adversarial constraints for pseudo-label generation.

Main Results:

• The ACCL-CINet achieved high-precision pathological image segmentation.
• Experimental results on public and private datasets demonstrated superior performance compared to state-of-the-art semi-supervised methods.
• The proposed method effectively utilizes unlabeled data for improved segmentation accuracy.

Conclusions:

• The ACCL-CINet presents a robust and effective solution for semi-supervised medical hyperspectral image segmentation.
• The adversarial consistency constraint learning strategy significantly enhances segmentation performance.
• The framework shows promise for advancing pathological image analysis.