A GROUNDBREAKING automated MRI segmentation model has been developed to provide accurate and robust segmentation of major anatomical structures across all MRI sequences. The new tool, named TotalSegmentator MRI, builds on the success of TotalSegmentator CT and addresses growing demand for an equivalent MRI solution.
Advancing Medical Imaging with AI
The study behind this innovation aimed to create a model capable of segmenting 80 anatomical structures, aiding clinical applications such as organ volumetry, disease characterisation, surgical planning, and opportunistic screening. Researchers trained the model using a combination of MRI and CT scans, ensuring its adaptability to real-world clinical settings.
Methodology and Evaluation
TotalSegmentator MRI was trained using 1,143 scans, comprising 616 MRI and 527 CT images, with a median patient age of 61 years. These were divided into a training set of 1,088 scans and an internal test set of 55 MRI scans. The model’s performance was also evaluated against two external datasets—AMOS and CHAOS—each containing 20 MRI scans, as well as against two publicly available models and the original TotalSegmentator CT tool.
Performance was assessed using Dice scores, a metric that measures the overlap between the AI-generated and expert radiologist segmentations. The new model achieved an impressive Dice score of 0.839 for 80 anatomical structures in the internal test set, surpassing the two alternative models (Dice score of 0.862 vs. 0.759 for 40 structures and 0.838 vs. 0.560 for 13 structures; P < .001 for both comparisons). When tested on the TotalSegmentator CT dataset (89 CT scans), its performance closely matched that of the original CT model (Dice score 0.966 vs. 0.970; P < .001).
Clinical Impact and Future Applications
The model was further tested on a dataset of 8,672 abdominal MRI scans to study age-related volume changes. Results revealed a strong correlation between age and organ volume, such as a negative correlation between liver volume and age (ρ = −0.096; P < .0001). These findings underscore the potential for AI-driven tools in medical research and diagnostics.
With its open-source availability and easy-to-use interface, TotalSegmentator MRI represents a major advancement in medical imaging, extending the capabilities of automated segmentation to MRI scans from any sequence. The tool is expected to significantly improve efficiency in radiology and clinical workflows, enhancing patient care through precise anatomical analysis.
Victoria Antoniou, EMJ
Reference
D’Antonoli TA et al. TotalSegmentator MRI: Robust Sequence-independent Segmentation of Multiple Anatomic Structures in MRI. Radiology. 2025;314(2):e241613.
