Author: Matthew C Abramowitz, Alan Dal Pra, Rodrigo Delgadillo, Nesrin Dogan, John C. Ford, Kyle R. Padgett, Levent Sensoy, Benjamin Spieler, Matthew T. Studenski, Jace Allen Walker 👨🔬
Affiliation: University of Miami, Department of Radiation Oncology, University of Miami, University of Miami Sylvester Comprehensive Cancer Center, University of Miami School of Medicine 🌍
Purpose:
Toxicities that affect a patient’s quality-of-life due to prostate cancer (pCa) radiation therapy (RT) are receiving more attention as RT has become increasingly successful in treating pCA. Previously, radiomics models of pCA relied on radiomic features (RF) extracted from a single modality to predict changes in international prostate symptom scores (delta-IPSS). The objectives of this work are to explore the performance of mixed modality models (MMM) and their single modality model (SMM) counterparts to predict delta-IPSS using RF, and to assess whether MMM improves model predictive performance.
Methods:
This study included twenty-one patients treated with volumetric arc therapy for pCa. Radiation oncologists with expertise in pCA delineated prostate contours on CT with the aid of diagnostic MRI and the Prostate Imaging Reporting and Data System guidelines. Then, the contours were transferred to T2-weighted MRI and first fraction CBCT. Forty-two RF were extracted from the T2, CT, and CBCT images. RF were pre-selected for each modality independently. Pre-selection was performed by ranking RF importance using random forest. Logistic regression models for delta-IPSS were generated using the selected DRF. The area under curve (AUC) of these models were calculated using 1000-iteration bootstrapping. Mixed modality models (MMM) were also generated using a combination of the pre-selected features for various pairs of modalities.
Results:
The mean AUC for the SMM of delta-IPSS were 0.81 for T2, 0.72 for CT, and 0.7 for CBCT. Whereas the mean AUC for MMM of delta-IPSS were 0.85 for T2+CT, 0.81 for T2+CBCT, and 0.79 for CT+CBCT. While the T2 SMM had the highest mean AUC, the T2+CT resulted in a higher performing ∆IPSS model.
Conclusion:
Overall, the MMM performed better than their SMM counterparts. Thus, future radiomic studies may benefit from the use of multiple modalities in the prediction of quality-of-life indicators for pCA patients.