Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces PatchICL, a hierarchical in-context learning framework for medical image segmentation that uses selective image patching to focus on informative anatomical regions. PatchICL employs multi-level supervision to train the model to actively identify and attend to these regions, reducing computational redundancy. Experiments show that PatchICL achieves competitive in-domain CT segmentation accuracy with 44% less compute than a global-attention baseline and outperforms it on several out-of-domain datasets, especially those with localized pathology.
Achieve comparable segmentation accuracy with nearly half the compute by intelligently focusing on informative image patches, rather than relying on brute-force global attention.
In-context learning (ICL) enables medical image segmentation models to adapt to new anatomical structures from limited examples, reducing the clinical annotation burden. However, standard ICL methods typically rely on dense, global cross-attention, which scales poorly with image resolution. While recent approaches have introduced localized attention mechanisms, they often lack explicit supervision on the selection process, leading to redundant computation in non-informative regions. We propose PatchICL, a hierarchical framework that combines selective image patching with multi-level supervision. Our approach learns to actively identify and attend only to the most informative anatomical regions. Compared to UniverSeg, a strong global-attention baseline, PatchICL achieves competitive in-domain CT segmentation accuracy while reducing compute by 44\% at $512\times512$ resolution. On 35 out-of-domain datasets spanning diverse imaging modalities, PatchICL outperforms the baseline on 6 of 13 modality categories, with particular strength on modalities dominated by localized pathology such as OCT and dermoscopy. Training and evaluation code are available at https://github.com/tidiane-camaret/ic_segmentation