Radiomics models based on automatic segmentation of pretreatment apparent diffusion coefficient maps showed promise for predicting the biochemical recurrence of advanced prostate cancer in a recent study.
Many prostate cancer (PCa) patients experience biochemical recurrence (BCR) during treatment. Early BCR detection is essential for timely intervention to limit disease progression and improve survival. Radiomics is a method based on extracting quantitative features from radiology images and using the data to make a clinical decision support system.
A study in the Journal of Applied Clinical Medical Physics aimed to develop radiomics models based on automatic segmentation of pretreatment MRI apparent diffusion coefficient (ADC) maps to predict advanced PCa BCR after radiotherapy, hormonal therapy, or other systemic therapy.
Patient Characteristics
A total of 100 patients were enrolled and randomly assigned to training (n=70) and testing (n=30) groups for the models. Of these, 35 were BCR-positive (n=25 in the training cohort, n=10 in the test cohort). No significant differences were observed in the age and clinical characteristics between the training and test datasets.
Development of the Radiomics Models
Features were extracted from automatically segmented regions of interest on the ADC maps. Four predictive models were constructed and tested. Two predictive models were designed by combining age, serum prostate-specific antigen level, Gleason score, and pretreatment clinical staging with the prostate area (Model_1) or PCa area (Model_2). The other two predictive models were constructed based only on the automatic prostate area (Model_3) and automatic PCa area (Model_4).
Performance Evaluation of the Models
All the models were found to perform well, with relatively good accuracy, in predicting the BCR of advanced PCa from pretreatment ADC maps. The area under the curve (AUC) values were 0.770 (95% confidence interval (CI): 0.527–1.000), 0.793 (95% CI: 0.604–0.981), 0.840 (95% CI: 0.698–0.983), and 0.808 (95% CI: 0.627–0.988) for Model_1, Model_2, Model_3, and Model_4, respectively, in the test datasets. The radiomics-only models, i.e., Model_3 and Model_4, demonstrated slightly better performance in predicting advanced PCa BCR than the radiomics models combined with clinical characteristics, i.e., Model_1 and Model_2.
The predictive models based on the prostate area (Model_1 and Model_3) showed equivalent performance to those based on the PCa area (Model_2 and Model_4). The DeLong test demonstrated that there was no significant difference between each model (all p>0.05). Moreover, the 95% bootstrap confidence interval for the precision-recall area difference included 0 between each model (all p>0.05).
Hence, the radiomics models developed in the study were predictive of advanced PCa BCR. Their predictive information may aid in the selection of appropriate, individualized treatment.
Source:
Wang, H., Wang, K., Ma, S., Gao, G., & Wang, X. (2023). Investigation of radiomics models for predicting biochemical recurrence of advanced prostate cancer on pretreatment MR ADC maps based on automatic image segmentation. Journal of Applied Clinical Medical Physics. https://doi.org/10.1002/acm2.14244
