Is Complete Labeling Necessary? Understanding Active Learning in Longitudinal Medical Imaging

2025-11-22

AI Summary

Longitudinal medical image change detection faces challenges including high annotation cost and frequent omission of subtle lesions. Existing active learning methods are ill-suited for multi-temporal difference identification tasks. To address this, we propose LMI-AL—the first deep active learning framework tailored for longitudinal change detection. LMI-AL generates baseline–follow-up image pairs via deformable image registration and slice-wise differencing, then introduces a dual-criterion sample selection strategy that jointly leverages differential saliency and model uncertainty to guide iterative 2D slice-level annotation. This is the first systematic integration of active learning into longitudinal change detection. Empirical evaluation demonstrates that LMI-AL achieves comparable performance to fully supervised models using only ~8% of the annotated data, significantly improving both annotation efficiency and detection accuracy—particularly for subtle pathological changes.

Technology Category

• Machine Learning: Active Learning
• Computer Vision: Segmentation
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Detecting changes in longitudinal medical imaging using deep learning requires a substantial amount of accurately labeled data. However, labeling these images is notably more costly and time-consuming than labeling other image types, as it requires labeling across various time points, where new lesions can be minor, and subtle changes are easily missed. Deep Active Learning (DAL) has shown promise in minimizing labeling costs by selectively querying the most informative samples, but existing studies have primarily focused on static tasks like classification and segmentation. Consequently, the conventional DAL approach cannot be directly applied to change detection tasks, which involve identifying subtle differences across multiple images. In this study, we propose a novel DAL framework, named Longitudinal Medical Imaging Active Learning (LMI-AL), tailored specifically for longitudinal medical imaging. By pairing and differencing all 2D slices from baseline and follow-up 3D images, LMI-AL iteratively selects the most informative pairs for labeling using DAL, training a deep learning model with minimal manual annotation. Experimental results demonstrate that, with less than 8% of the data labeled, LMI-AL can achieve performance comparable to models trained on fully labeled datasets. We also provide a detailed analysis of the method's performance, as guidance for future research. The code is publicly available at https://github.com/HelenMa9998/Longitudinal_AL.

Problem

• Reducing labeling costs for longitudinal medical imaging change detection
• Addressing limitations of conventional active learning in dynamic imaging tasks
• Developing efficient annotation framework for subtle lesion change identification

Innovation

• Active learning framework for longitudinal medical imaging
• Selects informative slice pairs from baseline and follow-up scans
• Achieves full performance with under 8% labeled data