Open Radiology AI Model Released for CT & MRI Analysis
Radiology AI company Raidium has released Curia-2, an open source foundation model intended to serve as a shared technical framework for analyzing medical imaging data across radiology workflows. The model is designed to process both standard 2D images and volumetric 3D scans, with the goal of supporting automated detection of radiological findings and assisting clinical decision systems.
According to the developers, Curia-2 was evaluated across multiple benchmark tracks covering image analysis and finding detection tasks. The model reportedly achieved higher precision than several existing 2D radiology foundation models like MedGemma.
Image credit: 2D Track Results, Raidium
Curia-2 also reached 88.6% accuracy on 3D evaluation tracks, which the company positions as relevant for oncology imaging workflows where volumetric scans such as CT and MRI are common.
Image credit: 3D Track results, Raidium
The system is also described as more data-efficient during training, reportedly reaching stable performance using fewer training samples than comparable models. In addition, Raidium reports that Curia-2 can detect a wide range of radiological findings covering roughly 200 clinical entities using a vision-only architecture, suggesting that detailed structural analysis of medical images alone may approach the performance of vision-language models that rely on paired medical text to interpret imaging findings.
Raidium has released a 2-billion-parameter version of the model, Curia-2B, on Hugging Face, allowing researchers and developers to test the system and potentially adapt it for specific radiology applications. A larger version, Curia-2L can be accessed upon direct request to Raidium.
Foundation models in radiology are increasingly being explored as a base layer for multiple downstream tasks, including lesion detection, triage systems, structured report generation, and multimodal clinical decision support.
Just yesterday, FDA has granted 510(k) clearance to CT Brain, an AI triage system developed by Harrison.ai that analyzes routine non-contrast head CT scans to automatically flag signs of acute brain infarction and prioritize suspected stroke cases for faster radiologist review.
Topic: AI in Bio
