🆕 Today's Digest — Foundation Models in Medicine

New papers/posts collected 2026-07-03.

1 new today

📋 In brief

This batch spotlights a compelling architectural alternative to autoregressive generation in clinical NLP: discrete diffusion models applied to radiology reporting. The central finding is that a diffusion-based model can match or outperform a comparable autoregressive model on medical VQA while decoding substantially faster, challenging the assumption that autoregressive dominance extends to specialized medical domains. Beyond raw performance, the more clinically significant contribution is the bidirectional infill capability — the ability to complete arbitrary gaps within a partially written report — which opens the door to genuinely interactive, human-in-the-loop report drafting workflows that autoregressive models cannot natively support.

📝 Today's papers

Discrete Diffusion Language Models for Interactive Radiology Report Drafting

This paper adapts DiffusionGemma-26B, a discrete diffusion language model, for radiology report generation and benchmarks it against its autoregressive sibling Gemma-4-26B under identical LoRA finetuning. The diffusion model matches or exceeds the AR model on medical VQA while decoding 3.5–4.4× faster. Crucially, its bidirectional denoising enables any-order infill — filling arbitrary gaps in a report from fixed text on both sides — a capability structurally unavailable to autoregressive models.

📄 Table

Date	Paper	Venue
2026-07-01	Discrete Diffusion Language Models for Interactive Radiology Report Drafting	—