Researchers introduced MedicalMLOmni, a multi-agent AutoML framework for medical image analysis. It uses evidence-based planning to guide model selection and configuration. Multiple agents run parallel hyperparameter optimization concurrently. MedicalMLOmni works through a natural language interface. It removes coding requirements for domain experts. The framework achieved the best score on 19 of 20 benchmarks. PMC-VQA was the exception, where a vision-language model with broader pre-training performed better. Backbone-matched controls confirmed the gains came from the framework itself.
What changed
Medical AutoML required engineering expertise and was confined to narrow tasks like 3D segmentation.
What it unlocks
Domain experts with rich datasets but limited engineering expertise can now automate model training for diverse medical tasks.
- 19 of 20 benchmarks best score
- PMC-VQA exception
Sources