Radiomics-based prognosis classification for high-risk prostate cancer treated with radiotherapy.

Abstract

The present study aimed to investigate if CT-based radiomics features could correlate to the risk of metastatic progression in high-risk prostate cancer patients treated with radical RT and long-term androgen deprivation therapy (ADT). A total of 157patients were investigated and radiomics features extracted from the contrast-free treatment planning CT series. Three volumes were segmented: the prostate gland only (CTV_p), the prostate gland with seminal vesicles (CTV_psv), and the seminal vesicles only (CTV_sv). The patients were split into two subgroups of 100 and 57patients for training and validation. Five clinical and 62radiomics features were included in the analysis. Considering metastases-free survival (MFS) as an endpoint, the predictive model was used to identify the subgroups with favorable or unfavorable prognoses (separated by athreshold selected according to the Youden method). Pure clinical, pure radiomic, and combined predictive models were investigated. With amedian follow-up of 30.7months, the MFS at 1and 3years was 97.2% ± 1.5 and 92.1% ± 2.0, respectively. Univariate analysis identified seven potential predictors for MFS in the CTV_pgroup, 11in the CTV_psv group, and 9in the CTV_sv group. After elastic net reduction, these were 4predictors for MFS in the CTV_pgroup (positive lymph nodes, Gleason score, H_Skewness, and NGLDM_Contrast), 5in the CTV_psv group (positive lymph nodes, Gleason score, H_Skewnesss, Shape_Surface, and NGLDM_Contrast), and 6in the CTV_sv group (positive lymph nodes, Gleason score, H_Kurtosis, GLCM_Correlation, GLRLM_LRHGE, and GLZLM_SZLGE). The patients' group of the training and validation cohorts were stratified into favorable and unfavorable prognosis subgroups. For the combined model, for CTV_p,the mean MFS was 134 ± 14.5 vs. 96.9 ± 22.2months for the favorable and unfavorable subgroups, respectively, and 136.5 ± 14.6 vs. 70.5 ± 4.3months for CTV_psv and 150.0 ± 4.2 vs. 91.1 ± 8.6months for CTV_sv, respectively. Radiomic features were able to predict the risk of metastatic progression in high-risk prostate cancer. Combining the radiomic features and clinical characteristics can classify high-risk patients into favorable and unfavorable prognostic groups.