University of Crete (UOC)
Radiomics in shoulder ultrasound: applications of novel image analysis and artificial intelligence methods for the detection and characterization of joint pathology
Abstract
dc:descriptionShoulder disorders are a major cause of pain and disability and they are among the most common indications for musculoskeletal imaging. Ultrasound plays a pivotal role in the evaluation and management of shoulder pathology due to its widespread availability, high spatial resolution, dynamic assessment capability and use in image-guided interventions. However ultrasound interpretation remains largely qualitative and may not fully capture the biological heterogeneity of shoulder diseases. Radiomics, a quantitative imaging methodology that extracts large numbers of image-derived features, has emerged as a promising method for converting medical images into objective biomarkers suitable for prediction, characterization, and clinical decision support. The aim of this thesis was to investigate the role of ultrasound radiomics in shoulder pathology, with particular emphasis on calcific tendinopathy and rotator cuff disease. Four complementary studies were conducted. The first study evaluated whether radiomics extracted from ultrasound images could predict technical success following ultrasound-guided percutaneous irrigation of calcific tendinopathy (US-PICT). Machine learning models incorporating radiomic and clinical features outperformed models based solely on clinical or radiomics variables alone, while external validation confirmed the generalizability of the proposed approach. The second study examined the influence of image preprocessing on radiomic feature reproducibility. Advanced contrast-limited adaptive histogram equalization (CLAHE) improved feature stability and increased the number of highly reproducible radiomic features, highlighting the importance of the preprocessing standardization. The third study investigated the potential prediction of one-year functional outcome following US-PICT and demonstrated that the combined clinical-radiomic model provided good predictive performance. Calibration curve and decision curve analysis were also included to assess model calibration and clinical utility. Furthermore, SHAP (SHapley Additive exPlanations) analyses were generated for the final models to facilitate interpretation of each of the selected feature contribution to model predictions. Finally, a prospective paired MRI-Ultrasound study explored the relationship between radiomics extracted from matched tendon footprint regions. MRI-derived radiomic features demonstrated stronger associations with rotator cuff pathology severity, whereas ultrasound-derived features showed limited discriminatory capability and minimal cross-modal correspondence. The findings of this thesis suggest that ultrasound radiomics can provide clinically relevant information and improve outcome prediction in selected applications, particularly in calcific tendinopathy of the rotator cuff. At the same time, the results emphasize the importance of reproducibility and methodological standardization. Ultrasound and MRI appear to generate distinct radiomic representations of musculoskeletal pathology, indicating that quantitative imaging biomarkers are likely to be modality- and pathology-specific rather than universally transferable. These findings contribute to the growing field of musculoskeletal radiomics and support the continued development of robust quantitative imaging biomarkers for personalized patient care.
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Triantafyllou, Matthaios
- Τριανταφύλλου, Ματθαίος
Subjects
dc:subject × 14- Ραδιομική
- Υπερηχοτομογραφία ώμου
- Μηχανική μάθηση
- Radiomics
- Shoulder Ultrasound
- Machine learning
- Κλινική Ιατρική
- Ιατρική και Επιστήμες Υγείας
- Ραδιολογία, Πυρηνική ιατρική και Απεικονιστική
- Ιατρική Βιοτεχνολογία
- Clinical Medicine
- Medical and Health Sciences
- Radiology, Nuclear Medicine and Imaging
- Medical Biotechnology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62072