Basel University Hospital presents TotalSegmentator: a deep learning segmentation model that can automatically segment important anatomical structures in body CT images

The number of CT scans performed and the data processing capacity available have increased in recent years. Thanks to advances in deep learning approaches, the capability of image analysis algorithms has been greatly improved. As a result of improvements in data storage, processing speed, and quality of algorithms, larger sample sizes have been used in radiological research. Segmentation of anatomical structures is crucial for many of these investigations. Radiological image segmentation can be used for advanced biomarker extraction, automatic pathology detection, and tumor burden quantification. Segmentation is already used in common clinical analyzes for purposes such as surgery and radiation planning.

Separate models exist for segmenting individual organs (such as the pancreas, spleen, colon, or lung) in CT images, and there has also been research on combining data from multiple anatomical structures into a single model. However, all of the above models include only a small subset of essential anatomical structures and are trained on small data sets that are not representative of routine clinical images. The lack of accessibility to many segmentation models and data sets severely limits their usefulness to researchers. Accessing publicly available data sets often requires extensive paperwork or requires the use of data providers that are difficult to work with or limited in pace.

Researchers at the Clinic for Radiology and Nuclear Medicine at the University Hospital Basel used around 1,204 CT data sets to create a method for segmenting 104 anatomical entities. They acquired the data set with CT scanners, acquisition settings, and contrast phases. Their model, TotalSegmentator, can segment most anatomically important structures in the body with minimal user intervention, and does so reliably in any clinical setting. The high accuracy (Dice score of 0.943) and robustness to multiple clinical data sets make this tool superior to others freely available online. The team also used a huge data set of more than 4,000 CT examinations to examine and report age-related volume and attenuation changes in various organs.

The researchers have made their model available as a pre-trained Python package for anyone to use. They highlight that since their model uses less than 12 GB of RAM and does not need a GPU, it can be run on any standard computer. Its dataset is also easily accessible and requires no special permissions or requests to download. The current research used a model based on nnU-Net because it has been shown to produce reliable results on various tasks. It is currently considered the gold standard for medical frame segmentation, outperforming most other approaches. Hyperparameter tuning and investigation of different models such as transformers improve the performance of the nnU-Net standard.

As mentioned in their article, the proposed model has several possible uses. In addition to its obvious surgical applications, rapid and easily accessible organ segmentation allows individual dosimetry, as has been demonstrated in the case of the liver and kidneys. Additionally, automated segmentation can improve research by providing clinicians with normal or even age-dependent parameters (HU, volume, etc.). Coupled with a lesion detection model, his model could be used to approximate tumor burden in a given body part. Additionally, the model can serve as a basis for developing models designed to identify various diseases.

The model has been downloaded by more than 4,500 researchers for use in various contexts. Only recently has it become possible to analyze data sets of this size, and it required a lot of time and effort on the part of data scientists. This work has demonstrated associations between age 12 years and the volume of numerous segmented organs using a data set of more than 4,000 people who had undergone multiple trauma CT scans. Common literature figures on normal organ size and age-dependent organ growth are usually based on sample sizes of a few hundred people.

The team highlights that male patients were overrepresented in the study’s data sets, which may be because, on average, more men than women visit hospitals. However, the team believes that their model can be a starting point for more extensive investigations of radiological populations. They mention that future studies will include more anatomical structures in their data set and model. Additionally, they are recruiting additional patients, adjusting for potential confounders, and performing more correlation analyzes to conduct a more complete study of aging.

Review the Paper and Project. All credit for this research goes to the researchers of this project. Also, don’t forget to join. our 33k+ ML SubReddit, 41k+ Facebook community, Discord channel, and Electronic newsletterwhere we share the latest news on ai research, interesting ai projects and more.

If you like our work, you’ll love our newsletter.

we are also in Telegram and WhatsApp.

<!– ai CONTENT END 2 –>