Development of AI to predict the prognosis of cervical spinal cord surgery...94% accuracy
Dec 18, 2024
|
Professor Gong Hyun-joong of the Department of Convergence Medicine at Seoul National University Hospital (researcher Seo Ye-chan, trainee Jeong Jeong-yi) and Professor Kim Chi-heon of Neurosurgery announced that they have developed an AI algorithm to track clinical information such as the recovery of nerve function of patients who underwent surgery for cervical spinal cord spondylosis from July 2015 to April 2017 and predict the prognosis of long-term surgery based on this.
Cervical spondylosis is a disease in which nerves in the cervical spine (neckbone) are damaged under pressure, and can cause motor and sensory nerve paralysis, such as difficult hand movement or difficulty walking. For treatment, the area where the spinal nerve passes (the laminoplasty) is opened and the pressure is relieved 'Cervical laminoplasty' is performed. After this operation, all patients should receive regular outpatient treatment once every few months for 1 year after surgery and once every year from 2 years to follow up on their prognosis.
However, patients who quickly recover their neural function and keep it stable may not require regular outpatient care. In fact, patients with such a good prognosis do not feel uncomfortable, so there is also a problem of no-show in outpatient treatment. In order to improve patient convenience and increase the efficiency of hospital operation, it was necessary to select patients with good prognosis in advance so that a customized follow-up schedule could be established.
The research team developed an AI model that predicts long-term prognosis based on postoperative clinical information of 80 patients with cervical spinal cord spondylosis. The model is designed to analyze variables such as gender, age, BMI, complications, and preoperative nerve function using machine learning techniques and to predict the recovery of nerve function two years after surgery. The patient's neurological function was measured by the widely used 'JOA score (the score of the Japanese Orthopedic Society)', and a score of 14.25 out of 17 was considered to have a good prognosis because the neurological function was well recovered.
Performance analysis showed that the sensitivity, specificity, and accuracy of the machine learning model were 97%, 88%, and 94%, respectively. In particular, the AUROC (under-curve area) level is 0.90, which means that patients with good prognosis and those who do not can be distinguished with high accuracy.
In addition, analysis of the effect of each variable on the prediction by the SHAP technique showed that 'JOA score' at 12 months of surgery had the most significant effect. In addition, ▲JOA score at 6 months, 1 month, and 3 months of surgery ▲ BMI ▲ preoperative JOA score ▲ gender ▲ age ▲ MRI snake eye sign ▲ preoperative walking status, etc. had a significant impact.
Additionally, we used data from 22 patients who underwent surgery from September 2020 to July 2022 to validate the model's performance and showed consistent outstanding performance with AUROC 0.86.
Professor Kim Chi-heon (neurosurgery) explained, "This study is significant because it presents the possibility of efficiently operating the cervical spondylosis treatment system. Patients with good prognosis predicted by AI can be treated on an optimized schedule, such as reducing the frequency of hospital visits, which will help improve the patient experience."
The findings were published in the recent issue of the international journal 『BMC Medical Informatics and Decision Making』.
|
bellho@sportschosun.com
