Machine learning-based construction and validation of an radiomics model for predicting ISUP grading in prostate cancer: a multicenter radiomics study based on [68Ga]Ga-PSMA PET/CT.
Authors
Affiliations (9)
Affiliations (9)
- Department of Urology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, No.87 Xiangya Road, Changsha City, Hunan Province, China.
- Department of Urology, Cheeloo College of Medicine, Qilu Hospital, Shandong University, Jinan, China.
- Magnetic Field-free Medicine & Functional Imaging, Research Institute of Shandong University, Jinan, China.
- Department of Nuclear Medicine, Affiliated Hospital of Qingdao University, Qingdao, 266000, China.
- Department of PET Center, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China.
- Department of Pathology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China. [email protected].
- Department of Urology, Affiliated Hospital of Qingdao University, Qingdao, 266000, China. [email protected].
- Department of Urology, Cheeloo College of Medicine, Qilu Hospital, Shandong University, Jinan, China. [email protected].
- Department of Urology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, No.87 Xiangya Road, Changsha City, Hunan Province, China. [email protected].
Abstract
The International Society of Urological Pathology (ISUP) grading of prostate cancer (PCa) is a crucial factor in the management and treatment planning for PCa patients. An accurate and non-invasive assessment of the ISUP grading group could significantly improve biopsy decisions and treatment planning. The use of PSMA-PET/CT radiomics for predicting ISUP has not been widely studied. The aim of this study is to investigate the role of <sup>68</sup>Ga-PSMA PET/CT radiomics in predicting the ISUP grading of primary PCa. This study included 415 PCa patients who underwent <sup>68</sup>Ga-PSMA PET/CT scans before prostate biopsy or radical prostatectomy. Patients were from three centers: Xiangya Hospital, Central South University (252 cases), Qilu Hospital of Shandong University (External Validation 1, 108 cases), and Qingdao University Medical College (External Validation 2, 55 cases). Xiangya Hospital cases were split into training and testing groups (1:1 ratio), with the other centers serving as external validation groups. Feature selection was performed using Minimum Redundancy Maximum Relevance (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) algorithms. Eight machine learning classifiers were trained and tested with ten-fold cross-validation. Sensitivity, specificity, and AUC were calculated for each model. Additionally, we combined the radiomic features with maximum Standardized Uptake Value (SUVmax) and prostate-specific antigen (PSA) to create prediction models and tested the corresponding performances. The best-performing model in the Xiangya Hospital training cohort achieved an AUC of 0.868 (sensitivity 72.7%, specificity 96.0%). Similar trends were seen in the testing cohort and external validation centers (AUCs: 0.860, 0.827, and 0.812). After incorporating PSA and SUVmax, a more robust model was developed, achieving an AUC of 0.892 (sensitivity 77.9%, specificity 96.0%) in the training group. This study established and validated a radiomics model based on <sup>68</sup>Ga-PSMA PET/CT, offering an accurate, non-invasive method for predicting ISUP grades in prostate cancer. A multicenter design with external validation ensured the model's robustness and broad applicability. This is the largest study to date on PSMA radiomics for predicting ISUP grades. Notably, integrating SUVmax and PSA metrics with radiomic features significantly improved prediction accuracy, providing new insights and tools for personalized diagnosis and treatment.