Radiomic analysis in the assessment of supraspinatus muscle tissue status on magnetic resonance imaging: a retrospective study

INTRODUCTION: Radiomics provides quantitative image-derived descriptors that may reduce subjectivity in MRI assessment of supraspinatus muscle degeneration. In rotator cuff pathology, the major challenge is reliable stratification of intermediate grades of fatty infiltration and atrophy, where visual grading is prone to inter-reader variability. Proton-density fat-suppressed sequences are routinely acquired in shoulder MRI and may serve as a practical source for radiomics-based modeling. 

OBJECTIVE: To identify the most informative radiomic, predominantly texture-based, features on proton-density fat-suppressed MRI of the supraspinatus muscle and, based on these features, to develop and internally validate a Random Forest machinelearning model for objective multiclass stratification of degenerative changes.

MATERIALS AND METHODS: A retrospective dataset of 41 shoulder MRI examinations acquired on 1.5 T scanners in multiple clinical sites was analyzed. PD-weighted turbo/fast spin-echo fat-suppressed series were selected and processed using a unified pre-processing and standardization workflow. The supraspinatus muscle was manually segmented slice-by-slice to create a three-dimensional region of interest. Radiomic features were extracted using SlicerRadiomics, followed by stepwise dimensionality reduction and a combined feature-selection strategy to form the final predictor set. A Random Forest classifier with class balancing and hyperparameter tuning was used as the main model. Performance was assessed with 5-fold stratified cross-validation using accuracy, macro F1-score, Cohen’s kappa, per-class sensitivity, and one-vs-rest ROC AUC.

Statistics: statistical data processing and the construction of machine learning models were carried out using IBM SPSS Statistics software (IBM Corp., Armonk, USA) and the R statistical computing environment (R Core Team, 2025), as well as using the Python programming language (scikit-learn, pandas, matplotlib, shap libraries).

RESULTS: Expert grading yielded an imbalanced class distribution (grade 0: 11 cases; grade 1: 5; grade 2: 20; grade 3: 5). The final modeling set comprised 43 radiomic predictors. Cross-validated performance of the Random Forest model reached an overall accuracy of 63.4%, macro F1-score of 45.7%, and Cohen’s kappa of 0.42. One-vs-rest ROC analysis showed high AUC values (0.94–1.0), while class-wise sensitivity indicated pronounced heterogeneity, with better recognition of the most prevalent category and reduced performance for intermediate/rare grades. Wavelet-based texture features and intensity-related descriptors contributed most to model decisions.

CONCLUSIONS: The study demonstrates the feasibility of radiomics-assisted multiclass classification of supraspinatus muscle degeneration on PD fat-suppressed MRI in a real-world clinical dataset. Texture features, particularly wavelet-derived metrics, were among the most informative; however, discrimination of adjacent and underrepresented grades remains a limitation and requires further protocol harmonization and larger datasets.

1. Peeters N.H.C., Kraats A.M., Krieken T.E. et al. The validity of ultrasound and shear wave elastography to assess the quality of the rotator cuff // Eur. Radiol. 2024. Vol. 34. P. 1971–1978. doi: 10.1007/s00330-023-10037-z.

2. Werthel J.D., Boux C.F., Burdin V. et al. CT-based volumetric assessment of rotator cuff muscle in shoulder arthroplasty preoperative planning // Bone Joint Open 2021. Vol. 2, No. 7. P. 552–561. doi: 10.1302/2633-1462.27.BJO-2021-0081.R1.

3. Nozaki T., Tasaki A., Horiuchi S. et al. Quantification of Fatty Degeneration Within the Supraspinatus Muscle Using a 2-Point Dixon Method on 3-T MRI // AJR Am. J. Roentgenol. 2015. Vol. 205, No. 1. P. 116–122. doi: 10.2214/AJR.14.13518.

4. Kissenberth M.J., Rulewicz G.J., Hamilton S.C. et al. A positive tangent sign predicts the repairability of rotator cuff tears // J. Shoulder Elbow Surg. 2014. Vol. 23, No. 7. P. 1023–1027. doi: 10.1016/j.jse.2014.02.014.

5. Ro K., Kim J.Y., Park H. et al. Deep-learning framework and computer-assisted fatty infiltration analysis for the supraspinatus muscle in MRI // Sci Rep. 2021. Vol. 11. Р. 15065. doi: 10.1038/s41598-021-93026-w.

6. Kim J.Y., Ro K., You S. et al. Development of an automatic muscle atrophy measuring algorithm to calculate the ratio of supraspinatus in supraspinous fossa using deep learning // Comput. Methods Programs Biomed. 2019. Vol. 182. Р. 105063. doi: 10.1016/j.cmpb.2019.105063.

7. Fuchs B., Weishaupt D., Zanetti M. et al. Fatty degeneration of the muscles of the rotator cuff: assessment by computed tomography versus magnetic resonance imaging // J. Shoulder Elbow Surg. 1999. Vol. 8, No. 6. P. 599–605. doi: 10.1058/S1058-2746(99)90097-6.

8. Fei Y., Wan Y., Xu L. et al. Novel methods to diagnose rotator cuff tear and predict post-operative Re-tear: Radiomics models // Asia Pac. J. Sports Med. Arthrosc. Rehabil Technol. 2024. Vol. 37. P. 14–20. doi: 10.1016/j.asmart.2024.03.003.

9. Saavedra J.P., Droppelmann G., García N., Jorquera C., Feijoo F. High-accuracy detection of supraspinatus fatty infiltration in shoulder MRI using convolutional neural network algorithms // Front Med (Lausanne). 2023. Vol. 10. Р. 1070499. doi: 10.3389/fmed.2023.1070499.

10. Ingebritsen R., Grogan B., Baer G., Cotter E. Deep learning accurately predicts supraspinatus pathology on shoulder magnetic resonance imaging // Arthrosc Sports Med. Rehabil. 2023. Vol. 5, No. 4. P. e1107–e1113. doi: 10.1016/j.asmr.2023.100807.

11. Vasilev Y.A., Nanova O.G., Blokhin I.A. et al. Priority radiomic parameters for computed tomography of head and neck malignancies: A systematic review // Digital Diagnostics. 2024. Vol. 5, No. 2. P. 255–268. doi: 10.17816/DD623240.

12. Xie J., Zhou M., Guo Z., Zhu Y., Jiang C.A. Quantitative Fatty Infiltration Evaluation of the Supraspinatus Muscle: Enhanced Clinical Relevance and Improved Diagnostic Value on Predicting Retear Compared With the Goutallier Classification // Am. J. Sports Med. 2025. Vol. 53, No. 4. P. 952–960. doi: 10.1177/036354652412313809.

13. Tian Z., Ni Y., He H. et al. Quantitative assessment of rotator cuff injuries using synthetic MRI and IDEAL-IQ imaging techniques // Heliyon. 2024. Vol. 10, No. 17. Р. e37307. doi: 10.1016/j.heliyon.2024.e37307.

14. Chianca V., Bottino V., Cuocolo R., Zappia M. MRI Quantitative Evaluation of Muscle Fatty Infiltration // Magnetochemistry. 2023. Vol. 9, No. 4. Art. 111. doi: 10.3390/magnetochemistry9040111.

15. Cheng C., Liang X., Guo D., Xie D. Application of Artificial Intelligence in Shoulder Pathology // Diagnostics (Basel). 2024. Vol. 14, No. 11. Р. 1091. doi: 10.3390/diagnostics14111091.

16. Rodriguez H.C., Rust B., Hansen P.Y. et al. Artificial Intelligence and Machine Learning in Rotator Cuff Tears // Sports Med. Arthrosc. Rev. 2023. Vol. 31, No. 3. P. 67–72. doi: 10.1097/JSA.0000000000000371.

17. Khoruzhaya A.N., Ahkmad E.S., Semenov D.S. The role of the quality control system for diagnostics of oncological diseases in radiomics // Digital Diagnostics. 2021. Vol. 2, No. 2. P. 170–184. doi: 10.17816/DD60393.

18. Mercuri E., Pichiecchio A., Allsop J. et al. Muscle MRI in inherited neuromuscular disorders: past, present, and future // J. Magn. Reson. Imaging. 2007. Vol. 25, No. 2. P. 433–440. doi: 10.1002/jmri.20804.

19. Sudoł-Szopinska I., Jacques T., Gietka P., Cotten A. Imaging in dermatomyositis in adults and children // J. Ultrason. 2020. Vol. 20. No. 80. P. e36–e42. doi: 10.15557/JoU.2020.0007.

20. Zhan J., Liu S., Dong C. et al. Shoulder MRI-based radiomics for diagnosis and severity staging assessment of surgically treated supraspinatus tendon tears // Eur. Radiol. 2023. Vol. 33, No. 8. P. 5587–5593. doi: 10.1007/s00330-023-09523-1.

21. Mannil M., Burgstaller J.M., Held U. et al. Correlation of texture analysis of paraspinal musculature on MRI with different clinical endpoints: Lumbar Stenosis Outcome Study (LSOS) // Eur. Radiol. 2019. Vol. 29, No. 1. P. 22–30. doi: 10.1007/s00330-018-5552-6.

22. Larue R.T., Timmeren J.E., Jong E.E.C. et al. Influence of gray level discretization on radiomic feature stability for different CT scanners, tube currents and slice thicknesses: a comprehensive phantom study // Acta Oncol. 2017. Vol. 56, No. 11. P. 1544–1553. doi: 10.1080/0284186X.2017.1351624.

23. Shi H., Duan Y., Shi J. et al. Role of preoperative prediction of microvascular invasion in hepatocellular carcinoma based on the texture of FDG PET image: A comparison of quantitative metabolic parameters and MRI // Front Physiol. 2022. Vol. 13. Р. 928969. doi: 10.3389/fphys.2022.928969.

24. Aerts H.J., Velazquez E.R., Leijenaar R.T. et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach // Nat. Commun. 2014. Vol. 5. Р. 4006. doi: 10.1038/ncomms5006.

25. Gillies R.J., Kinahan P.E., Hricak H. Radiomics: images are more than pictures, they are data // Radiology. 2016. Vol. 278, No. 2. P. 563–577. doi: 10.1148/radiol.2015151169.

26. Lee D., Hong K.-T., Lee W. et al. Threshold-based quantification of fatty degeneration in the supraspinatus muscle on MRI as an alternative method to Goutallier classification and single-voxel MR spectroscopy // BMC Musculoskelet Disord. 2020. Vol. 21, No. 1. Р. 362. doi: 10.1186/s12891-020-03400-4.

27. Danilov G.V., Agrba S.B., Strunina Yu.V. et al. Radiomics and Machine Learning in Diagnostics of Glial Brain Tumors: a Systematic Review and Meta-Analysis // Sovrem. Tekhnologii Med. 2025. Vol. 17, No. 6. P. 69–79. doi: 10.17691/stm2025.17.6.07.