Distance Learning-Based Prototypical Network With Multi-Domain Adaptation for Few-Shot Hyperspectral Medical Image...

Area of Science:

• Medical imaging
• Machine learning
• Biomedical engineering

Background:

• Hyperspectral imaging (HSI) offers potential for medical diagnostics by analyzing spectral signatures.
• Challenges include limited labeled data for training and domain shift across datasets, hindering generalization.

Purpose of the Study:

• To develop a novel distance-learning-based prototypical network for few-shot hyperspectral medical image classification.
• To address data scarcity and domain shift issues in clinical HSI analysis.

Main Methods:

• Proposed a class-covariance-aware Mahalanobis metric within a prototypical network to adapt similarity measures.
• Introduced a domain-aware adapter block for dynamic fusion of spectral-spatial representations and domain-specific characteristics.
• Validated on skin dermoscopy, choledochal, and in-vivo brain HSI datasets.

Main Results:

• The novel method demonstrated enhanced prototype robustness under label scarcity.
• Significantly reduced misclassification compared to existing few-shot networks.
• Achieved excellent performance across three diverse medical HSI datasets.

Conclusions:

• The proposed method effectively overcomes limitations of label scarcity and domain shift in medical HSI classification.
• Paves the way for generalizable HSI solutions in clinical workflows and biomedical research.
• Highlights the potential of advanced machine learning for precise disease detection using HSI.