AUTHOR=Xie Jun , Yang Yi , Jiang Zekun , Zhang Kerui , Zhang Xiang , Lin Yuheng , Shen Yiwei , Jia Xuehai , Liu Hao , Yang Shaofen , Jiang Yang , Ma Litai TITLE=MRI radiomics-based decision support tool for a personalized classification of cervical disc degeneration: a two-center study JOURNAL=Frontiers in Physiology VOLUME=Volume 14 - 2023 YEAR=2024 URL=https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2023.1281506 DOI=10.3389/fphys.2023.1281506 ISSN=1664-042X ABSTRACT=Objectives: To develop and validate an MRI radiomics-based decision support tool for automated grading cervical disc degeneration.The retrospective study included 435 patients with 2610 cervical disc samples from two hospitals. The cervical magnetic resonance imaging (MRI) analysis of patients confirmed cervical disc degeneration grades using the Pfirrmann grading system. A training set (305 patients with 1830 samples) and an independent test set (130 patients with 780 samples) were divided for the machine learning model construction and validation, respectively. We provided a fine-tuned MedSAM model for cervical disc automatically segmentation. Then, we extracted 924 radiomics features from each segmented disc in T1 and T2 MRI. All features were processed and selected by using minimum redundancy maximum relevance and multiple machine learning algorithms. Meanwhile, the radiomic models of various machine learning algorithms and MRI images were constructed and compared.Finally, the combined radiomic model was built in the training set and validated in the test set.Radiomics feature mapping was provided for auxiliary diagnosis.Results: Of the 2610 cervical disc samples, 794 (30.4%) were defined as low-grade and 1816 (69.6%) were high-grade. The fine-tuned MedSAM model achieved good segmentation performance, with the mean dice of 0.93. Higher order texture features contributed the dominant force in diagnostic task (80%). Through comparisons of various machine learning models, random forest showed the higher performance than other algorithms (P < 0.01); and T2 MRI radiomic model achieved better than T1 MRI in the diagnostic performance (P < 0.05). The final combined radiomic model had an area under the receiver operator characteristic curve (AUC) of 0.95, an accuracy of 89.51%, a precision of 87.07%, a recall of 98.83%, and an F1 score of 0.93 in the test set, which all were better than other models (P < 0.05).The radiomics-based decision support tool using T1 and T2 MRI can be used for cervical disc degeneration grading, facilitating individualized management.