2025 ISAKOS Biennial Congress Paper
Trustworthy Deep Learning for the Automated Quantification of Fatty Infiltration of the Rotator Cuff Muscles Using MRI
Asma Salhi, PhD, Brisbane, QLD AUSTRALIA
Ruth Delaney, FRCS , Dublin IRELAND
Freek Hollman, MD, PhD, Sint-Michielsgestel NETHERLANDS
Kristine R. Italia, MD, FPOA, Quezon City, Metro Manila PHILIPPINES
Sarah L Whitehouse, PhD, Brisbane, Queensland AUSTRALIA
Yuantong Gu, Prof, Brisbane, QLD AUSTRALIA
Kenneth Cutbush, MBBS, FRACS, FAOrthA, Spring Hill, QLD AUSTRALIA
Ashish Gupta, MBBS, MSc, FRACS, FAORTHOA, Brisbane, QLD AUSTRALIA
Quasr Collaborative, (QUASR), Brisbane, QLD AUSTRALIA
Queensland Unit for Advanced Shoulder Research (QUASR), Brisbane, QLD, AUSTRALIA
FDA Status Not Applicable
Summary
This study aims to sssess the effective utilisation of artificial intelligence advancements for the automated measurement of the fatty infiltration of the RC muscles using shoulder MRI images.
Abstract
Introduction
Fatty infiltration of the rotator cuff (RC) muscles is a commonly associated with various shoulder disorders, particularly with rotator cuff tears. Accurate and efficient quantification of fatty infiltration is essential for treatment planning, decision-making, and monitoring disease progression. However, the current method of classifying fatty infiltration is highly subjective and has low reliability. In this study, we propose a new deep-learning (DL) approach to automatically classify fatty infiltration using a proposed simplified Goutallier classification system from MRI images.
Methods
MRI scans of patients without or with varying degrees of cuff tears and chronicity of injury were included in this study. A medial scapular body (MSB) Goutallier classification was utilised to classify the fatty infiltration of the RC muscles. This data was independently evaluated by nine subspecialists (six fellowship-trained shoulder surgeons and three musculoskeletal radiologists) to assess the reliability and validity of the Goutallier classification for grading the fatty infiltration. The four muscles were classified and grossly segmented after determining the main sections following this method. 1149 images of segmented RC muscles were used to train the AI models. A novel DL pipeline comprising key components of in-domain transfer learning, feature fusion, and machine learning classifiers was proposed to classify RC fatty infiltration automatically. Pre-trained DL models Xception, InceptionV3, and MobileNetV2 were trained separately. Then, K-Nearest Neighbour (KNN), Support Vector Machines (SVM), and Naive Bayes classifiers were trained using fused features extracted by three DL models from the delineated RC muscle areas.
Results
The assessment of individual DL models showed that MobileNetV2 demonstrated the highest overall performance, achieving an accuracy of 89.5%, a specificity of 94.7%, a recall of 89.5%, a precision of 90.5%, and an F1-score of 90.0%. The experimental results of the feature fusion demonstrated that combining these components has excellent potential to achieve reliable, robust, and enhanced classification outcomes. Among the classifiers used, KNN achieved the highest performance, with an accuracy of 91.1%, a specificity of 95.5%, a recall of 91.1%, a precision of 93.1%, and an F1-score of 92.1%. Furthermore, the use of Grad-CAMs confirmed that the networks had learned relevant clinical features in the region of interest.
Conclusion
This research provides evidence for the effective utilisation of artificial intelligence advancements for the automated measurement of the fatty infiltration of the RC muscles using shoulder MRI images. The accuracy of this approach is similar to that of subspecialists. Additionally, the study showcases the potential of using ensembles of deep classifiers that are fine-tuned with transfer learning. This approach improved accuracy and Grad-CAMs that were consistent with clinical knowledge. Providing automated measurement of fatty infiltration can assist clinicians with their treatment approach and decision-making for patients with RC tears.