Newest papers

Expandable Modality Experts for Medical Vision-Language Learning

DM BAPPY, Eunsu Kim, Hanjoo Lee, Lee Sh, Sangjeong Ahn · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29881 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Aprender através de modalidades de imagem médica requer o uso de imagens sem texto emparelhado e acomodar novos domínios, mantendo as representações existentes. Proporcionamos uma técnica de visão médica construída em torno de um banco expandível de especialistas em modalidade. Seu Multiway Transformer combina computação visual compartilhada com especialistas de feed-forward deterministicamente roteados e uma tomografia de texto separada em pares. Um tokenizador visual consciente da modalidade fornece alvos para pré-treinamento mascarado em 19,4 milhões de linhas de imagem não emparelhadas em sete modalidades; o alinhamento proposto por contraste usa então 3,23 milhões de pares de imagem-texto. Fundos adicionais e especialistas em tomografia de coerência óptica aprendem a partir de dados emparelhados, enquanto os caminhos

Machine translation (Meta M2M100); the English original follows.

Abstract

Learning across medical imaging modalities requires making use of images without paired text and accommodating new domains while retaining existing representations. We propose a medical vision-language technique built around an expandable bank of modality experts. Its Multiway Transformer combines shared visual computation with deterministically routed feed-forward experts and a separate text pathway. A modality-aware visual tokenizer supplies targets for masked pretraining on 19.4 million unpaired image rows across seven modalities; contrastive alignment then uses 3.23 million image-text pairs. Additional fundus and optical coherence tomography experts learn from paired data while the original image and text embedding paths remain frozen. With a ViT-Large backbone, the proposed technique achieves the best or tied-best result in 24 of 25 benchmark settings across nine modalities: 10 of 11 zero-shot classification benchmarks, all 10 linear probes, and all four retrieval benchmarks. Mean balanced accuracy is 0.674 for zero-shot classification and 0.856 for linear probing, and mean bidirectional retrieval recall is 20.70%. A separate three-seed sequential experiment with matched adaptation data yields fundus and OCT retrieval gains of 14.95 and 9.31 percentage points, with no measured loss on preceding fundus or pathology tasks. The preservation argument requires unchanged existing computation paths and inference conditions, and feature-identity checks verify retention on evaluated inputs. The proposed technique provides a modular approach to learning and expanding medical vision-language representations.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Artificial intelligence, Medical imaging, Modular design, Modality (human–computer interaction), Encoder

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange