Darsy X-Ray Analysis

Multimodal generative AI for radiological analysis — detection, anatomical segmentation, and GAN-based image enhancement, trained with federated learning.

Darsy is Unite4AI’s flagship healthcare model, built on federated-learning research that lets hospitals collaborate on model training without ever pooling patient data.

What it does

Progressive Growing GAN preprocessing — enhances degraded, over/under-exposed, or artifact-heavy X-ray, CT, and MRI images before analysis.
Diagnostic detection — flags abnormalities across conditions including cancers, cardiovascular disease, and diabetic retinopathy, with calibrated confidence scores.
Anatomical segmentation — distinguishes tissue, bone, organs, and lesions, localizing the extent of detected findings.
Continuous monitoring — combines detection and segmentation for longitudinal tracking and disease-progression forecasting.

Why federated

Medical privacy regulations rightly prevent centralizing patient data, and no single hospital has enough data to train a strong model alone. Federated learning resolves the deadlock: the model travels to the data, gradients travel back, raw records never leave the institution. Differential privacy is applied on top of the aggregation step.

Intended use and limits

Not a medical device. This model is a research and decision-support tool. It is not cleared by any regulator for autonomous diagnosis. Every output requires review by a qualified clinician.

Known limitations: performance varies across scanner manufacturers and imaging protocols; evaluation cohorts under-represent several populations — see the evaluation report before deploying in any new setting.