arXiv 2602.158112026-05-25cs.CVcs.AI

CARL-CXR：基于连续适配器路由的任务未知胸部X光片分类

CARL-CXR: Continual Adapter-Based Routing for Task-Unknown Chest Radiograph Classification

Department of Imaging Physics, The University of Texas MD Anderson Cancer Center（影像物理系，德克萨斯大学MD安德森癌症中心）

Muthu Subash Kavitha, Anas Zafar, Amgad Muneer, Jia Wu

备注：: 9 pages, 4 figures

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

提出CARL-CXR框架，通过固定高容量骨干网络、增量添加轻量级任务特定适配器和分类头，以及潜在任务选择器，解决任务未知推理下的胸部X光片增量分类问题，显著减少灾难性遗忘并提升路由准确性。

摘要：

胸部X光片分类器的临床部署需要模型能够在新数据集可用时进行更新，而无需对先前观察到的数据进行重新训练或降低已验证的性能。我们研究了任务未知推理下的任务增量连续学习设置，其中异质的胸部X光数据集顺序到达，且在部署时任务身份不可用。我们提出了CARL-CXR，一个基于连续适配器的路由框架，该框架保持固定的高容量骨干网络，同时增量引入轻量级任务特定适配器和分类头。一个潜在任务选择器基于适配器条件特征进行操作，将每个输入动态路由到最相关的任务路径，利用紧凑的任务原型和特征级经验回放来在顺序更新中保留任务身份，而无需存储原始图像。在MIMIC-CXR和CheXpert两个具有不同患者群体、成像设备和注释流程的大规模数据集上的实验表明，CARL-CXR实现了最小的灾难性遗忘（AUROC下降0.012），比已建立的连续学习基线LwF和EWC分别减少了6倍和11倍，同时保持了具有竞争力的诊断性能（AUROC 0.74）。在任务未知部署下，CARL-CXR在路由准确性上比联合训练高出12.5个百分点（75.0% vs. 62.5%）：与LwF和EWC不同，后者在推理时需要明确的任务标识符且不提供路由机制。

英文摘要：

Clinical deployment of chest radiograph classifiers requires models that can be updated as new datasets become available without retraining on previously observed data or degrading validated performance. We study a task-incremental continual learning setting for chest radiograph classification under task-unknown inference, where heterogeneous chest X-ray datasets arrive sequentially and task identity is unavailable at deployment time. We propose CARL-CXR, a continual adapter-based routing framework that maintains a fixed high-capacity backbone while incrementally introducing lightweight task-specific adapters and classifier heads. A latent task selector operates on adapter-conditioned features to dynamically route each input to the most relevant task pathway, leveraging compact task prototypes and feature-level experience replay to preserve task identity across sequential updates without storing raw images. Experiments on MIMIC-CXR and CheXpert two large-scale datasets with distinct patient populations, imaging devices, and annotation pipelines demonstrate that CARL-CXR achieves minimal catastrophic forgetting (0.012 AUROC drop), representing a 6X and 11X reduction over established continual learning baselines LwF and EWC respectively, while maintaining competitive diagnostic performance (AUROC 0.74). Under task unknown deployment, CARL-CXR outperforms joint training by 12.5 points in routing accuracy (75.0% vs. 62.5%): unlike LwF and EWC, which require explicit task identifiers at inference and provide no routing mechanism.

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。