Summary

Radiomics is gaining relevance in the medical field. This study assessed the statistical dependence between preoperative MRI radiomic features and pre-and post-operative PROMs in patients who underwent arthroscopic rotator cuff repair. The radiomics features most frequently related to PROMs were Maxi- mum2DDiameterColumn, Flatness, LeastAxisLength, GLCM-Correlation, GLCM-MCC, and GLCM-Imc2.

Abstract

Background

Rotator cuff tears (RCTs) are the most common shoulder musculoskeletal disorders. Generally, clinicians base their surgical decisions on the evaluation of MRI scans, specifically assessing the condition of the tendon, the size and extent of the tear, and signs of muscle atrophy. While Patient Reported Outcome Measures (PROMs) capture patients’ personal experiences, a more objective and accurate analysis of MRI can contribute to more precise decision-making. Identifying radiomics features correlating with the subjective scores from PROMSs could offer a comprehensive and objective assessment of a patient’s overall health status, which could be instrumental for clinicians in managing RCTs.
AIMS
We hypothesize that radiomics features from MRI scans may offer a more objective and measurable indicator of supraspinatus tears than PROMs. This study assessed the statistical dependence between preoperative MRI radiomics features and pre- and post-operative PROMs.

Methods

In this study, 44 patients (20 females, 24 males, mean age 58.9 ± 7.8) who underwent arthroscopic rotator cuff repair for full-thickness RCTs of any grade between January 2019 and January 2022 were enrolled. RCTs were classified according to Cofield and Gerber's classification. Patients evaluations were performed the day before surgery, and at 1, 3, 6, and 12 months after surgery. SPADI, SST, ASES, OSS, CMS, HADS, SF-36 were administered at each follow-up. Coronal, T2-weighted spin-echo MRI scans were obtained in all patients and evaluated independently by two blinded orthopedic surgeons. Two complementary approaches were employed to assess the discriminative and predictive relevance of radiomics features with each PROM, i.e., the F-statistics was used for evaluating the strength of linear relationship, while the non-linear interactions were analyzed through a model-free AI-based Mutual Information estimator. Radiomics features were extracted using an Imaging Biomarker Standardization Initiative (IBSI)-compliant open-source software. The 107 extracted features can be broadly categorized into 18 intensity-based features, 14 shape-based features, and five texture matrices from which a total of 75 texture-based features were extracted.

Results

In both preoperative and postoperative analysis, the radiomics features most frequently related to the patient’s health condition were three shape-based features, namely the maximum diameter on the coronal plane (Maximum2DDiameterColumn), the Flatness, and the smallest axis length of the region of interest (ROI)-enclosing ellipsoid (LeastAxisLength), and three texture-based features (Gray Level Co-occurrences Matrix descriptors, GLCM), namely, the spatial correlation between gray-level values (i.e., GLCM-correlation), the Maximal Correlation Coefficient (GLCM-MCC), and the Informational Measure of Correlation 2 (GLCM-Imc2).

Discussion

One of the major findings was the predominance of the shape-based feature class, as well as a remarkable relevance of the GLCM characteristics when assessing the severity of the preoperative clinical condition; moreover, we found that the evolution of the overall clinical picture was driven by a more heterogeneous set of visual features, with particular emphasis on a wider ensemble of texture descriptors. All the identified features showed a high significance through both analytical methods, indicating a consistent influence on various clinical outcomes. Incorporating F-statistics and Mutual Information in the analyses adds depth to the findings, establishing a strong groundwork for future advancements in the field.

Acknowledgement: Funded by the European Union - Next Generation EU - NRRP M6C2 - Investment 2.1 Enhancement and strengthening of biomedical research in the NHS (Project no. PNRR-MAD-2022-12376080 - CUP: F83C22002450001).