FIGURE 9 from A Pilot Study on Patient-specific Computational Forecasting of Prostate Cancer Growth during Active Surveillance Using an Imaging-informed Biomechanistic Model

Patient-specific forecasts of prostate cancer risk. A–D, Time trajectories of the model-based markers involved in the calculation of our prostate cancer risk classifier (left) as well as the trajectory of the latter (right) for 4 patients. The patients shown in this figure are the same considered in Figs. 4 and 6. In all panels, darker hues represent results from the global calibration scenario (GC, see Fig. 4), where the model is fit to the three mpMRI datasets from each patient, while lighter hues show results from the fitting-forecasting scenario (FF, see Fig. 6), where the model is only fit to the first two mpMRI datasets from each patient. Dotted gray vertical lines in the background indicate the times of the mpMRI scans for each patient. Solid lines correspond to quantities calculated from the model fit, while dashed lines correspond to values calculated from model forecasts. The model-based markers of interest are the total tumor index (NT, green curves, left vertical axis) and mean proliferation activity of the tumor (Ap, pink curves, right vertical axis). The prostate cancer risk classifier was trained with the global calibration results (see Fig. 8), yielding an optimal performance threshold that separates lower risk prostate cancer (blue region) from higher-risk prostate cancer (red region). The prostate cancer risk at the times of histopathologic assessment of the tumors (i.e., biopsy, surgery) is represented as a bullet point, and the corresponding GS values are annotated over the prostate cancer risk trajectory from the global calibration scenario. In A, the patient exhibits a low-risk tumor during AS, which is correctly identified by our classifier in both computational scenarios. In B, our model consistently classifies the patient's tumor as higher-risk during the majority of AS in both computational scenarios. In C, the patient initially exhibits a lower-risk tumor that had progressed to higher-risk at the time of surgery (i.e., terminal GS value). In the fitting-forecasting scenario our model consistently predicts that the tumor is a higher-risk case, while, in the global calibration scenario, tumor progression is detected shortly after the second mpMRI scan. Importantly, in the fitting forecasting scenario, our approach identifies a higher-risk tumor 1,011 days earlier than standard practice (i.e., assessment of the surgical specimen). In D, our model consistently identifies the tumor as a higher-risk case in both computational scenarios.