Monday, December 2 | 1:50 p.m.-2:00 p.m. | M6-SSMK03-3 | E450A
In this scientific presentation, researchers will share how their radiomics algorithm can predict both the onset and progression of knee osteoarthritis (OA).
Huimao Zhang, MD, of The First Hospital of Jilin University, China, and colleagues developed their model using data from 456 participants in the Osteoarthritis Initiative (OAI). All patients received 3-tesla MRI of the right knee, with radiomic features extracted from cartilage and subchondral bone. The result is a set of MRI-based logistic regression models cascaded into a "one-stop" multiclassification system for predicting knee OA onset and progression, respectively, according to the presenters.
For their test of model performance in relation to knee OA occurrence, onset, and progression, Zhang and colleagues prospectively stratified participants into OA and non-OA groups based on four-year radiographic follow-up and 1:2 propensity score matching. Early and late progressors were further classified based on the timing of knee OA progression.
"The novel cascade radiomic pipeline we constructed to predict the occurrence and progression of [knee OA] can provide one-stop technical support for early diagnosis of [knee OA], screening of high-risk populations, and prediction of disease progression," the team said.
They noted that the combined cartilage and subchondral bone radiomics models demonstrated superior performance in predicting onset and progression, reporting an AUC of 0.985 for predicting knee OA occurrence and 0.738 for progression, for example.
Of particular interest, the model predicting progression of knee OA based on subchondral bone performed at an AUC of 0.721, slightly higher than that of the cartilage-based model where AUC was 0.676. However, after cascading the LR models, the team predicted AUC values of each category to be above 0.8 with an accuracy of 78.1%
Learn more about their results by attending this presentation.