Promptable Longitudinal Lesion Segmentation in Whole-Body CT

2025-08-30

AI Summary

Longitudinal lesion segmentation in whole-body CT suffers from poor temporal consistency across timepoints. Method: This paper proposes a promptable longitudinal lesion segmentation framework built upon an extended LongiSeg architecture, integrating a Transformer-based segmentation backbone with interactive point/mask prompting mechanisms and large-scale synthetic-data-driven self-supervised pretraining to enhance modeling of longitudinal contextual dependencies. Contribution/Results: The method enables efficient transfer learning with minimal annotated data, achieving up to a 6-percentage-point improvement in Dice score over from-scratch training baselines on real clinical datasets. It significantly improves temporal consistency, robustness, and clinical interpretability of lesion tracking across timepoints, thereby providing a reliable technical foundation for quantitative assessment of disease progression.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Data transparency and provenance

Abstract

Accurate segmentation of lesions in longitudinal whole-body CT is essential for monitoring disease progression and treatment response. While automated methods benefit from incorporating longitudinal information, they remain limited in their ability to consistently track individual lesions across time. Task 2 of the autoPET/CT IV Challenge addresses this by providing lesion localizations and baseline delineations, framing the problem as longitudinal promptable segmentation. In this work, we extend the recently proposed LongiSeg framework with promptable capabilities, enabling lesion-specific tracking through point and mask interactions. To address the limited size of the provided training set, we leverage large-scale pretraining on a synthetic longitudinal CT dataset. Our experiments show that pretraining substantially improves the ability to exploit longitudinal context, yielding an improvement of up to 6 Dice points compared to models trained from scratch. These findings demonstrate the effectiveness of combining longitudinal context with interactive prompting for robust lesion tracking. Code is publicly available at https://github.com/MIC-DKFZ/LongiSeg/tree/autoPET.

Problem

• Accurate longitudinal lesion segmentation in whole-body CT
• Consistent tracking of individual lesions across time
• Leveraging limited training data with synthetic pretraining

Innovation

• Promptable segmentation with point and mask interactions
• Large-scale pretraining on synthetic CT dataset
• Longitudinal context integration for lesion tracking