Machine Learning-Based Ultrasomics for Predicting Subacromial Impingement Syndrome Stages.

Objectives: To determine the performance of machine learning (ML)-based ultrasomic analysis of subacromial impingement syndrome (SIS) stage evaluation.
Methods: In this retrospective study, 324 patients with SIS were included. The SIS stage was evaluated with a Neer test. Regions of the musculi supraspinatus were manually segmented by an experienced radiologist. Then, 5936 ultrasomic features were extracted from the Ultrasomics Platform software. The Wilcoxon test was used to identify differentially expressed radiomic features. Then, these differentially expressed features were submitted to the least absolute shrinkage and selection operator (LASSO) for model construction. The area under the curve (AUC) of the receiver operating characteristic was used to evaluate the performance of the ultrasonic model for SIS stage evaluation.
Results: Finally, a total of 223 early-stage and 101 advanced-stage SIS patients were randomly divided into a training cohort (n = 227) and a validation cohort (n = 97). After feature-dimensionality reduction, a total of 28 radiomic features were submitted to LASSO analysis. Finally, 10 radiomic features were finally included for radiomics model construction. The AUC results showed that the ultrasomics model had moderate performance for SIS stage evaluation in both the training cohort (AUC = 0.839) and the validation cohort (AUC = 0.789).
Conclusions: ML-derived ultrasomics can discriminate the SIS stage in patients with SIS. This noninvasive and low-cost approach may be helpful in the preliminary screening of shoulder pain.
(© 2021 American Institute of Ultrasound in Medicine.)