Medical Imaging Advances with State-of-the-Art AI Segmentation Model
Thanks to the development of deep learning approaches in artificial intelligence, there has been a significant advancement in the capability of image analysis algorithms. These algorithms are being used in radiological research to improve the segmentation of anatomical structures, which is vital for biomarker extraction, automatic pathology detection, and tumor load quantification.
The researchers at the Clinic of Radiology and Nuclear Medicine, University Hospital Basel, have developed a groundbreaking AI segmentation model called TotalSegmentator. This model leverages a large dataset of CT scans to accurately segment 104 anatomical entities with minimal user input. Their model has a high accuracy (Dice score of 0.943) and is robust across various clinical datasets, making it superior to others freely available online.
The TotalSegmentator is available as a pre-trained Python package, requiring less than 12 GB of RAM and no GPU, so it can run on any standard computer. Furthermore, their dataset is easily accessible, eliminating the need for special permissions or requests to download it. This accessibility addresses the limitations of previous models and datasets, making it a valuable tool for researchers.
In addition to surgical applications, the TotalSegmentator can be used for individual dosimetry, providing clinicians with normal or age-dependent parameters. It can also be used in conjunction with a lesion-detection model to approximate tumor loads and to identify various diseases. The model has already been downloaded by over 4,500 researchers for use in a variety of contexts.
The team at the University Hospital Basel is committed to further studies and improvements to their model. Future research will include more anatomical structures in their dataset and model, as well as recruiting additional patients to conduct a more comprehensive study of aging. This work has the potential to advance medical imaging and research, bringing us one step closer to developing life-saving technologies.