UMCG researcher develops AI software that equals radiologist in muscle and fat measurements

UMCG radiologist Alain Viddeleer has developed software that uses artificial intelligence to measure a person’s body composition, such as the distribution between muscle and fat tissue. The software uses images from a CT scan and does this equally well as an experienced radiologist.
Radioloog voor computerscherm

It takes a radiologist an average of five minutes per patient, while the system can analyse hundreds of patients at once. Overall muscle mass appears to be an independent measure of a person's fitness; patients with low muscle mass have a higher risk of complications during heavy operations. With the help of this new tool, this data can now be determined much faster and used in medical decisions.

Overall muscle mass as a measure of fitness

In recent years, increasing attention has been paid to the influence of body composition on various diseases. Previous research has shown that muscle mass is an independent measure of a person's condition. The condition of a person is an important factor for assessing the risks of operations. Patients with low muscle mass have a higher risk of complications during heavy operations. In addition, the distribution of body fat influences the course of many diseases, such as an infection with the coronavirus

Measuring muscle mass with a CT scan

A radiologist can use a CT scan to determine the muscle mass in the body. The CT scan makes an image of a cross-section of the body, a so-called "slice" on which the muscle and fat tissue can be clearly seen. Radiologists now determine the muscle mass by manually drawing the 'slice' on the computer, so the system measures how much muscular mass is present within the outlined area. This takes about 5 minutes per patient and is very time-consuming for the radiologist. Moreover, in order to get accurate measurements, the drawing in must be done by specialised professionals.


AI software as accurate as an experienced radiologist

To speed up this process, radiologist Alain Viddeleer and his team have developed software that automates muscle mass determination with the help of artificial intelligence. The researchers trained this software with 10 million images from 7000 patients and examined how accurate the measurements were. They showed that the software can now measure the muscle mass of hundreds of patients within seconds with an accuracy of 99% compared to an experienced radiologist. The tool therefore offers the possibility of working much faster and of making the determination of muscle mass a standard measurement in CT scans. 

Insights in patients' fitness

Viddeleer is currently using the software for research purposes, but believes that in the future it can be used to support medical decisions. By gaining insight into the fitness of patients, doctors can better assess whether patients can cope with a heavy operation. In addition, this tool can be used to better inform patients about their body composition and the related lifestyle.