Researchers at Seoul National University Hospital (SNUH) have developed an artificial intelligence model to predict a patient’s sarcopenia with just a lower extremity X-ray and a blood test before conducting artificial knee joint replacement surgery.
Sarcopenia causes muscle mass and function to decrease with age. It is a leading risk factor for falls, fractures, and other postoperative complications.
Calf circumference figures are used to screen existing sarcopenia but are not recommended due to fat accumulation with age and changes in skin elasticity. Meanwhile, MRI or CT diagnosis has high muscle mass measurement accuracy, but is time-consuming, entails radioactive exposure, and obtains different measurement results depending on the observer.
Therefore, appropriate tools are needed to select sarcopenia with existing tests to avoid burdening patients with additional tests, according to a SNUH press release on Tuesday.
The research team led by Professor Ro Du-hyun used X-ray photographs of 227 healthy volunteers to estimate the total muscle mass of the patient by segmenting the muscles from X-ray photographs of the lower extremities. They also considered seven variables to predict sarcopenia, including systemic muscle mass, body mass index, bilirubin, hemoglobin, albumin, protein, and age.
The research team tested this model on 403 patients undergoing Total Knee Arthroplasty (TKA) for degenerative knee arthritis.
A specialist confirmed that the model showed high performance in automatic muscle segmentation of lower extremity X-ray photographs. Furthermore, the model’s area under the curve (AUC) value was 0.98, showing excellent prediction capability.
Among the seven variables predicting sarcopenia, the predicted mean value (PMV) of the systemic muscle mass was the most critical variable for determining sarcopenia.
The study confirmed that simple X-rays and blood tests can now diagnose sarcopenia before surgery instead of additional tests.
“By using this technology, we will be able to accurately predict sarcopenia of patients undergoing various orthopedic surgery and provide appropriate treatment,” Professor Ro said.
Hwang Doo-hyun, the study’s first author, said, “The research is a good example of directly developing technologies necessary for actual clinical practice by using a combination of deep learning and machine learning.”
The study was published in the latest issue of the International SCI Journal of Clinical Medicine.