分割
人工智能
计算机科学
自回归模型
图像分割
光学(聚焦)
模式识别(心理学)
计算机视觉
组分(热力学)
机器学习
尺度空间分割
医学影像学
合成数据
可视化
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-12-15
卷期号:: 4125-4130
标识
DOI:10.1109/bibm66473.2025.11356940
摘要
Medical images contain implicit temporal information that reflects disease progression. Predicting future lesion development from existing imaging data is essential for proactive clinical decision-making. However, most retinal segmentation methods focus on static analysis and fail to capture the dynamic nature of retinal pathologies. To address this, we propose a predictive segmentation framework based on spatiotemporal mapping, integrating latent space conditional diffusion with a pathology-guided evolution mechanism. The method is grounded in the hypothesis that spatial transitions between adjacent OCT slices approximate temporal pathological progression, enabling temporal modeling from spatial cues. An autoregressive strategy with time-series modeling is used to infer future lesion changes, while a structural control mechanism ensures anatomical realism and individual consistency. Experimental results show that the proposed model effectively simulates disease trajectories, maintains structural integrity, and produces clinically meaningful predictions. This work provides a promising direction for time-aware medical image analysis and early intervention in retinal diseases.
科研通智能强力驱动
Strongly Powered by AbleSci AI