ESTRO 2024 - Abstract Book

S5078

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

We used PET and mpMRI data to differentiate between high-risk and low-risk lesions that were defined in terms of immunoreactivity scores for Ki67, PSA or a combination of the two. The cut-offs for high-risk lesions were based on percentiles of the cohort, similarly to Hammarsten et al 1 . High-risk lesions were either those with Ki67 ≥ Ki67 median (scenario 1), PSA ≤ PSA median (scenario 2) or those with both Ki67 ≥ 75th percentile and PSA ≤ 25th percentile (scenario 3). For the present cohort, the median Ki67 (11%) was close to the 10% threshold used in similar works [2–4]. Low-risk lesions were defined as those with Ki67 < Ki67 median in scenario 1, and PSA > PSA median in scenario 2. In scenario 3, the low-risk lesions were defined as those having both Ki67 ≤ 25th percentile and PSA ≥ 25th percentile. PET was quantified by maximum standardized uptake values (SUV max ). mpMRI was quantified by the maximum volume transfer constant ( N K trans ), fifth percentile apparent diffusion coefficient ( N ADC 5% ) and fifth percentile T2w image intensity ( N T2 5% ) (each normalized to non-malignant prostatic tissue, signified by the subscript N). The Mann-Whitney U test was used to compare image summary measures between groups of lesions. Logistic regression models were fitted with one or several image summary measures as individual variables. Performances were evaluated using receiver operating characteristics (ROC) analysis, and the area under the ROC curves (AUCs) were compared using two-tailed p-values. P-values < 0.05 were considered statistically significant. The evaluation resulted in 385 lesions. All image summary measures were significantly different between HR and LR lesions in scenario 1 and 3, and all except N T2 5% in scenario 2. As a single modality, PSMA-PET SUV max achieved the highest AUC in scenario 1 (AUC = 0.68) and scenario 3 (AUC = 0.81), whereas N K trans achieved the highest AUC in scenario 2 (AUC = 0.68). These best performing image summary measures were significant predictors of high-risk lesions after adjusting for the size of lesions (p < 0.01). Combining PSMA-PET SUV max with the mpMRI-based measures as individual variables in logistic regression models yielded highest AUCs overall (AUC = 0.71 in scenario 1 and 2, AUC = 0.85 in scenario 3). This was significantly higher than what was achieved by only combining the mpMRI-based models in scenario 1 (ΔAUC = 0.03, p < 0.05). The mpMRI-based models outperformed models only based on N ADC 5% and N T2 5% in scenario 2 (ΔAUC = 0.05, p < 0.05) and scenario 3 (ΔAUC = 0.08, p < 0.05). Fig. 1 illustrates the ROC curves for a selection of models in scenario 3, where biparametric MRI refers to combining N ADC 5% and N T2 5% as individual variables logistic regression models. Results: