arXiv 2602.235332026-02-26eess.IVcs.CVcs.LG

小样本持续学习用于3D脑部MRI

Few-Shot Continual Learning for 3D Brain MRI with Frozen Foundation Models

Independent Researcher, USA（美国独立研究者）；Indiana University at Bloomington, United States（印第安纳大学布卢明顿分校）；Independent Researcher, Taiwan（台湾独立研究者）；National Taiwan University, Taiwan（台湾国立大学）；Boise State University, United States（博伊西州立大学）；National Yang Ming Chiao Tung University, Taiwan（台湾阳明交通大学）

Chi-Sheng Chen, Xinyu Zhang, Guan-Ying Chen, Qiuzhe Xie, Fan Zhang, En-Jui Kuo

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出基于冻结基础模型与任务特定LoRA模块的小样本持续学习方法，有效解决3D脑部MRI肿瘤分割与脑年龄估计任务中的灾难性遗忘问题。

摘要：

预训练在大规模3D医学影像数据上的基础模型在持续学习中面临适应多个下游任务时受限于少量标记数据的挑战。我们通过结合冻结预训练的主干网络与任务特定的低秩适应（LoRA）模块，解决3D脑部MRI的小样本持续学习问题。任务按顺序到来——肿瘤分割（BraTS）和脑年龄估计（IXI）——没有先前任务数据的回放。每个任务接收一个专用的LoRA适配器；只有适配器和任务特定的头部被训练，而主干网络保持冻结，从而通过设计消除灾难性遗忘（BWT=0）。在持续学习中，顺序全微调导致严重遗忘（T1 Dice从0.80降至0.16在T2后），而顺序线性探测在T1（Dice 0.79）表现强劲但T2失败（MAE 1.45）。我们的LoRA方法在两个任务上实现了最佳平衡性能：T1 Dice 0.62±0.07，T2 MAE 0.16±0.05，无遗忘且每任务<0.1%可训练参数，尽管在T2中存在系统性的年龄低估（Wilcoxon p<0.001）。冻结基础模型与任务特定的LoRA适配器因此在需要维持两个任务的小样本持续学习中提供实用解决方案。

英文摘要：

Foundation models pretrained on large-scale 3D medical imaging data face challenges when adapted to multiple downstream tasks under continual learning with limited labeled data. We address few-shot continual learning for 3D brain MRI by combining a frozen pretrained backbone with task-specific Low-Rank Adaptation (LoRA) modules. Tasks arrive sequentially -- tumor segmentation (BraTS) and brain age estimation (IXI) -- with no replay of previous task data. Each task receives a dedicated LoRA adapter; only the adapter and task-specific head are trained while the backbone remains frozen, thereby eliminating catastrophic forgetting by design (BWT=0). In continual learning, sequential full fine-tuning suffers severe forgetting (T1 Dice drops from 0.80 to 0.16 after T2), while sequential linear probing achieves strong T1 (Dice 0.79) but fails on T2 (MAE 1.45). Our LoRA approach achieves the best balanced performance across both tasks: T1 Dice 0.62$\pm$0.07, T2 MAE 0.16$\pm$0.05, with zero forgetting and $<$0.1\% trainable parameters per task, though with noted systematic age underestimation in T2 (Wilcoxon $p<0.001$). Frozen foundation models with task-specific LoRA adapters thus offer a practical solution when both tasks must be maintained under few-shot continual learning.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。