Predictive Model of Acute Rectal Toxicity in Prostate Cancer Treated With Radiotherapy.

Area of Science:

• Radiation Oncology
• Medical Physics
• Oncology

Background:

• Acute rectal toxicity is a common side effect of radiotherapy for prostate cancer.
• Personalized treatment selection is crucial to minimize toxicity while maintaining efficacy.
• Current predictive models may not fully capture complex dosimetric factors across different radiation modalities.

Purpose of the Study:

• To develop and validate a predictive model for acute rectal toxicity in prostate cancer patients treated with photon and proton radiotherapy.
• To integrate dosimetric and non-dosimetric variables for improved prediction accuracy.
• To aid in personalized treatment selection for radiotherapy.

Main Methods:

• A prospective multi-institutional cohort of 278 patients was analyzed.
• A convolutional neural network (CNN) was developed to analyze dose-area histograms (DAHs) and integrate with non-dosimetric predictors.
• Model performance was compared against logistic regression (LR) using the area under the curve (AUC).

Main Results:

• Key predictors included rectum length, race, age, and hydrogel spacer use.
• The CNN model achieved an AUC of 0.81 in the validation set, outperforming LR (AUC = 0.54).
• The model showed high sensitivity and specificity in photon and proton therapy subsets, indicating strong predictive capability.

Conclusions:

• A novel approach effectively predicts rectal toxicity in prostate cancer patients undergoing radiotherapy with photons and protons.
• Integrating dosimetric features from DAHs using CNNs with non-dosimetric factors enhances predictive accuracy.
• The model shows potential for guiding treatment decisions and warrants prospective validation.