OCL: Ordinal Contrastive Learning for Imputating Features with Progressive Labels

2025-03-03

International Conference on Medical Image Computing and Computer-Assisted Intervention

AI Summary

Early diagnosis of Alzheimer’s disease (AD) faces challenges from incomplete multimodal neuroimaging data—due to high acquisition costs and participant burden—and ordinal, progressively deteriorating clinical labels. Method: We propose the first contrastive learning framework for ordinal missing-data imputation. It explicitly incorporates label ordinality into the contrastive objective via progressive pseudo-label distillation and ordinal-aware negative sampling, augmented by self-supervised reconstruction and graph-enhanced neighborhood consistency constraints. Contribution/Results: Unlike conventional imputation methods that ignore ordinal semantics, our approach ensures semantically consistent feature imputation. Evaluated on six real-world ordinal-missing benchmarks, it achieves an average 12.7% improvement in imputation accuracy and a 9.3% gain in downstream classification F1-score, significantly outperforming state-of-the-art imputation and ordinal learning methods.

Technology Category

• Machine Learning: Multimodal Learning
• Search and Optimization: Learning to Search
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Problem

• Imputing missing imaging data for Alzheimer's Disease stages.
• Leveraging diverse imaging modalities to retain all subjects.
• Enhancing alignment of imaging features across different modalities.

Innovation

• Ordinal contrastive loss for AD progression alignment
• Modality-independent embeddings via encoder-decoder networks
• Domain adversarial training for modality coherence