Figure 2.

Overview of the complete pipeline employed in this study. The process begins with image acquisition and retrospective data collection. Segmentation is performed next by individually processing each sacroiliac joint. Feature extraction follows, allowing us to retrieve a wide array of features from each segment. The data are then preprocessed, including a standardization step to normalize the range of feature values. The feature selection phase ensues, including using Pearson correlation and then the Boruta method to pinpoint the most significant features. The pipeline culminates in model training, where machine learning (ML) algorithms are trained on the selected features and used to classify each segment as either positive or negative for inflammatory BME (Created with BioRender [11], 18 June 2023).