Researchers in Switzerland have unveiled an innovative AI model called TotalSegmentator MRI, designed to automatically segment major anatomical structures in MRI images, which significantly alleviates the workload of radiologists.

Traditionally, MRI segmentation has been a manual, labor-intensive process that requires significant time and effort from radiologists, often leading to variability in results.

The TotalSegmentator MRI was trained on a comprehensive dataset comprising 616 MRI and 527 CT exams, enabling it to achieve sequence-independent segmentations of 80 anatomical structures.

Built on the nnU-Net framework, TotalSegmentator MRI is an open-source tool that adapts to various datasets with minimal user input, optimizing its performance for medical image segmentation.

In internal tests, the model achieved a Dice score of 0.839, indicating high accuracy in its segmentations, and it outperformed two other public models.

Dr. Jakob Wasserthal emphasized that TotalSegmentator MRI is unique in its ability to automatically segment the highest number of anatomical structures across various MRI sequences, enhancing precision in measurements.

The potential clinical applications of TotalSegmentator MRI include treatment planning, monitoring disease progression, and conducting opportunistic screening, according to Dr. Wasserthal.

The study detailing this AI model was published in the journal Radiology, showcasing its superior performance compared to other publicly available segmentation tools.

Dr. Wasserthal noted that automated systems like TotalSegmentator MRI can reduce human errors and provide more consistent results, ultimately improving diagnostic processes.

The innovation behind TotalSegmentator MRI lies in the extensive dataset utilized, which allows for effective segmentation across different MRI scanners and acquisition settings.

This model is a valuable asset for radiologists, capable of segmenting the highest number of anatomical structures from MRIs of any sequence.

TotalSegmentator MRI has already matched the performance of the related TotalSegmentator CT tool, which is utilized by over 300,000 users globally.