Prompt to Polyp: Medical Text-Conditioned Image Synthesis with Diffusion Models

Mikhail Chaichuk, Sushant Gautam, Steven Hicks, Elena Tutubalina · 2025

Read Prompt to Polyp: Medical Text-Conditioned Image Synthesis with Diffusion Models by Mikhail Chaichuk, Sushant Gautam, Steven Hicks, Elena Tutubalina (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-05573.txt.

Description

Mikhail Chaichuk, Sushant Gautam, Steven Hicks, and Elena Tutubalina ask whether a giant image model or a small specialist paints better medical pictures from words, in this 2025 paper. Synthetic medical images could relieve two chronic pains at once: too little training data for health AI, and privacy rules that lock real scans away. Text-to-image diffusion models promise both volume and variety.

Their specialist, MSDM, reworks Stable Diffusion with a clinical text encoder, a variational autoencoder, and cross-attention tuned to medical phrasing. It is compared against fine-tuned giants, FLUX and Kandinsky, on colonoscopy and radiology data sets. The giants won on raw fidelity. The small model delivered comparable quality at a fraction of the computing cost, a trade hospitals might gladly take. Medical experts joined the scoring, and the authors report plainly where each approach still fails, from anatomy errors to text misalignment.

Catalog record

CREATORMikhail Chaichuk, Sushant Gautam, Steven Hicks, Elena Tutubalina
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-05573

Browse the full library · About the Tower of Babel