# Prediction of Metastasis-Free Survival in Patients with Localized Prostate Adenocarcinoma Using Delta Radiomics from Pre-Treatment PSMA-PET/CT Scans and Dosiomics

## TL;DR

This study shows that combining pre- and post-treatment imaging data with clinical factors can better predict cancer recurrence in prostate cancer patients.

## Contribution

The novel integration of delta radiomics and dosiomics with clinical variables improves metastasis-free survival prediction in localized prostate cancer.

## Key findings

- Models using delta radiomics and clinical variables achieved a test c-score of 0.58 and AUC of 0.70.
- Dosiomics models with clinical variables achieved a test c-score of 0.56 and AUC of 0.67.
- Combined models outperformed clinical-only models in predicting metastasis-free survival.

## Abstract

This study develops prognostic models combining delta radiomics from PSMA-PET/CT, dosiomics, and clinical variables to predict metastasis-free survival (MFS) in patients with localized prostate adenocarcinoma treated with androgen deprivation therapy and external-beam radiotherapy. Delta radiomics features were computed from primary tumor volumes on pre- and post-treatment PSMA-PET/CT scans, while dosiomics features were derived from the intra-prostatic lesion receiving 86 Gy in the planning CT scans. Selected high-variance radiomics and dosiomics features were integrated with clinical factors, including age, Gleason score, baseline PSA, and PSA relapse. Data were split into training and testing cohorts with imbalance correction, and prognostic factors were evaluated using Cox regression and five-year MFS classification. Models incorporating delta radiomics or dosiomics with pre-treatment imaging and clinical variables consistently outperformed clinical factor-only models, achieving higher concordance indices and AUC values. These findings demonstrate that integrating biomarkers capturing temporal radiomics and spatial dose heterogeneity with clinical data improves prognostic accuracy and supports the use of radiomics and dosiomics as non-invasive tools for personalized radiotherapy planning in localized prostate cancer.

Purpose: To develop prognostic models integrating delta radiomics from prostate-specific membrane antigen positron emission tomography/computed tomography (PSMA-PET/CT) and dosiomics with clinical variables to predict metastasis-free survival (MFS) in patients with localized prostate adenocarcinoma treated with androgen deprivation therapy and external-beam radiotherapy. Materials/Methods: Delta-radiomics analysis included 43 patients. Radiomics features were extracted from the primary tumor on pre- and post-treatment PSMA-PET/CT, and delta features were calculated as relative changes. Eight high-variance features were selected and combined with clinical variables (age, Gleason score, initial PSA, and a binary variable, indicating the occurrence of PSA relapse). Data was split 70:30 with training-set imbalance correction. Predictors that were significant in univariate Cox regression (p < 0.05) were entered into multivariate Cox models, and five-year MFS was classified using a quadratic support vector machine. Dosiomics analysis included 48 patients. Dosiomics features were extracted from the planning target volume receiving 86 Gy and combined with pre-treatment radiomics and clinical variables using the same framework. Results: For delta radiomics, Model 1 (delta radiomics + pre-treatment radiomics + clinical) achieved the best performance (test c-score 0.58; AUC 0.70), exceeding Model 2 (pre-treatment radiomics + clinical; c-score 0.56; AUC 0.65) and Model 3 (clinical only; c-score 0.51; AUC 0.56). For dosiomics, Model 1 showed the highest performance (test c-score 0.56; AUC 0.67) compared with Model 2 (c-score 0.55; AUC 0.62) and Model 3 (c-score 0.50; AUC 0.54). Conclusions: Integrating delta radiomics or dosiomics with pre-treatment imaging and clinical variables improves MFS prediction and supports their role as non-invasive biomarkers for individualized radiotherapy in localized prostate cancer.