Ultrashort echo time MRI radiomics as a predictor of clinical outcomes in patellar tendinopathy: Insights from a large prospective clinical trial

Abstract

Purpose

To evaluate the predictive utility of radiomic features extracted from ultrashort echo time (UTE) MRI in comparison to conventional proton density (PD) sequences for short-term (24-week) and long-term (5-year) clinical outcomes in patients with patellar tendinopathy (PT) receiving exercise therapy.

Materials and methods

This prospective study of 76 PT patients undergoing 24-week exercise therapy underwent baseline 3D UTE and PD MRI at 3.0 T. The patellar tendon segmentation used nnU-Net, evaluated with Dice coefficient. Six predictive models consisting of clinical covariates and radiomic features from UTE and PD were developed using Elastic Net with 10-fold cross-validation. Model performance in predicting responsiveness of the patient-reported Victorian Institute of Sports Assessment (VISA-P) score was evaluated using the area under the receiver operating characteristic curve (ROC AUC) and the precision-recall curve (PR AUC), with 95% confidence intervals.

Results

The mean Dice similarity coefficient for the automatic segmentation of the patellar tendon from 3D-PD was 0.92 (SD: 0.02) and from 3D-UTE-Cones 0.89 (SD: 0.03). The UTE-based radiomics model demonstrated the highest predictive performance at 24 weeks (ROC AUC: 0.714 [95% CI: 0.701–0.727]; PR AUC: 0.848 [0.837–0.858]), while the PD-based model showed the lowest (ROC AUC: 0.569 [0.553–0.584]; PR AUC: 0.710 [0.692–0.727]). At the 5-year follow-up, UTE radiomics maintained robust performance (ROC AUC: 0.692 [0.677–0.706]; PR AUC: 0.822 [0.810–0.834]), whereas PD radiomics remained limited (ROC AUC: 0.578 [0.561–0.594]; PR AUC: 0.694 [0.676–0.713]).

Conclusions

Radiomics features extracted from UTE MRI demonstrate the highest predictive performance for clinical outcomes.