Deep Learning Approaches for Medical Imaging Under Varying Degrees of Label Availability: A Comprehensive Survey

2025-04-15

AI Summary

Deep learning in medical imaging critically depends on large-scale, high-quality annotated datasets—yet clinical labels are often scarce, imprecise, or entirely absent. To address this label scarcity challenge, this work systematically reviews approximately 600 peer-reviewed publications from 2018 onward. It introduces the first unified taxonomy distinguishing incomplete supervision, imprecise supervision, and unsupervised paradigms under label-constrained settings. We construct a cross-task–cross-modality relational map encompassing classification, segmentation, detection, and applications across brain, thoracic, and cardiac imaging. Furthermore, we propose an interpretable, mechanism-driven classification framework to comparatively evaluate the applicability and limitations of semi-supervised, weakly supervised, self-supervised, transfer learning, and generative annotation methods in medical contexts. This constitutes the most comprehensive methodology survey to date on label-constrained AI for medical imaging, offering practitioners an algorithm selection guide, clinical deployment recommendations, and a benchmark for future research.

Technology Category

• Computer Vision: Medical and Biological Imaging
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Deep learning has achieved significant breakthroughs in medical imaging, but these advancements are often dependent on large, well-annotated datasets. However, obtaining such datasets poses a significant challenge, as it requires time-consuming and labor-intensive annotations from medical experts. Consequently, there is growing interest in learning paradigms such as incomplete, inexact, and absent supervision, which are designed to operate under limited, inexact, or missing labels. This survey categorizes and reviews the evolving research in these areas, analyzing around 600 notable contributions since 2018. It covers tasks such as image classification, segmentation, and detection across various medical application areas, including but not limited to brain, chest, and cardiac imaging. We attempt to establish the relationships among existing research studies in related areas. We provide formal definitions of different learning paradigms and offer a comprehensive summary and interpretation of various learning mechanisms and strategies, aiding readers in better understanding the current research landscape and ideas. We also discuss potential future research challenges.

Problem

• Addressing medical imaging with limited labeled data availability
• Surveying deep learning methods for incomplete or inexact labels
• Analyzing 600+ studies on classification, segmentation, and detection tasks

Innovation

• Deep learning for medical imaging with limited labels
• Survey of incomplete inexact and absent supervision
• Covers classification segmentation detection tasks