PIKACHU: Prototypical In-context Knowledge Adaptation for Clinical Heterogeneous Usage

Amar Kumar, Zahra TehraniNasab, Emily Kaczmarek, Tal Arbel

Keywords: Foundation Models, In-Context Learning, Large Language Models, Out-of-distribution Generalization

Abstract: Medical imaging systems increasingly rely on large vision language foundation models (VLFMs) trained on diverse biomedical corpora, yet these models remain difficult to adapt to new clinical tasks without costly fine-tuning and large annotated datasets. We present PIKACHU (Prototypical In-Context Knowledge Adaptation for Clinical Heterogeneous Usage), a lightweight and generalizable framework that enables rapid few-shot adaptation of frozen medical FMs using only a handful of labelled examples. Unlike prior approaches that modify backbone weights or introduce heavy attention-based adapters, PIKACHU performs all task adaptation directly in the FM feature space through in-context proto-typical reasoning. Given a small support set, the framework constructs class prototypes by averaging normalized embeddings from a frozen VLFM image encoder and performs prediction on query images using temperature-scaled cosine similarity. Only a single temperature parameter is learned. We evaluate PIKACHU across three heterogeneous medical imaging datasets - dermatological images (ISIC), Optical Coherence Tomography (OCT), and Diabetic Retinopathy (DR), using established vision models (SigLIP, PubMedCLIP, DinoV2, and ViT) as backbones. The proposed in-context learning (ICL) strategy consistently outperforms the baseline (zero-shot) approaches across all datasets and architectures, achieving substantial improvements in both accuracy and AUC. Notably, with PubMed-CLIP as the backbone, PIKACHU achieves 0.69 accuracy on the ISIC dataset, 0.72 on OCT, and 0.79 on DR, demonstrating robust generalization across diverse clinical imaging modalities. These results highlight the promise of feature-space in-context learning as an efficient and deployable paradigm for test-time adaptation of foundation models, without the need for extensive retraining. To facilitate broader adoption and research, we make our code publicly available at https://github.com/Amarkr1/pikachu.

Primary Subject Area: Transfer Learning and Domain Adaptation

Secondary Subject Area: Foundation Models

Registration Requirement: Yes

Reproducibility: https://github.com/Amarkr1/pikachu

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 256