Periodic Reporting for period 2 - PICModForPCa (Personalised Image-based Computational Modelling Framework to Forecast Prostate Cancer)
Reporting period: 2022-09-01 to 2023-08-31
Prostate cancer (PCa) is a public health burden and a major concern among ageing men worldwide. Although the vast majority of PCa cases are currently detected at an early stage thanks to regular screening, their severity may vary wildly across patients. While a few prostatic tumors (<5%) are metastatic at diagnosis, about 75% cases of PCa show low to intermediate risk and may not produce any symptom nor require treatment for long time periods. However, nearly 85% of PCa patients receive an aggressive treatment (e.g. surgery or radiotherapy) with curative intent after diagnosis. Radical treatment for PCa can eliminate most diagnosed tumors, but the potential side-effects can also adversely impact the patient’s quality of life without prolonging longevity. Ideally, such intervention should be restricted to combat intermediate and high-risk PCa that may constitute a threat to the patient’s life. Additionally, some patients who initially receive no treatment may eventually exhibit rapid growth of aggressive tumors, experience treatment failure, and ultimately succumb to PCa due to an incorrect initial diagnosis. These issues result from the limited individualization of the clinical management of PCa and have led to significant rates of overtreatment and undertreatment. Alternatively, newly-diagnosed PCa patients with favorable pathological features may enroll in active surveillance (AS). In this clinical strategy, patients are closely monitored via multiparametric magnetic resonance imaging (mpMRI), PSA tests, and biopsies until these reveal an increase in PCa risk that warrants treatment. Thus, AS can contribute to reduce the current overtreatment of PCa. Additionally, the longitudinal clinical and imaging data collected for each patient provides key information about tumor growth that can inform a clinician in the selection of appropriate monitoring strategies or in the planning of treatments. Hence, AS also has potential to reduce the undertreatment of PCa patients. However, current AS protocols rely on monitoring a patient’s tumor according to observational population-based studies, which usually prescribe testing types (i.e. PSA, mpMRI, biopsies) at fixed times or following the suspicion of an increase in the risk level of PCa based on the collected data. This approach limits the design of personalized monitoring plans, complicates the early detection of tumor progression, and does not fully exploit the wealth of information about the patient’s tumor development in the longitudinal mpMRI and clinical data collected during AS. This project aims at addressing the current limitations of AS as well as the deficiencies and excesses in PCa treatment by leveraging patient-specific forecasts of PCa dynamics. Hence, the overarching goal of this project is to construct a personalized computational technology that integrates longitudinal clinical and mpMRI data into a predictive biomechanistic model of PCa that enables to run organ-scale simulations to improve diagnosis and obtain forecasts of PCa growth for each individual patient.
We have designed a biomechanistic model of PCa growth based on tumor cell density dynamics and relying on robust biological and biomechanical phenomena, which are represented by differential equations. This model is posed within the patient’s prostate geometry extracted from mpMRI data. The parameters of the model are identified on a patient-specific basis by solving an inverse problem using the patient’s longitudinal clinical and mpMRI data. The validation of the model consists of retrospectively comparing model predictions to further mpMRI and clinical data corresponding to dates that are posterior to the datasets leveraged in model calibration.
We performed a preliminary study in a small cohort of PCa cases (n = 7) who enrolled in AS at The University of Texas Health Science Center San Antonio (TX, USA), which we then validated in an ensuing study over a larger cohort over a larger PCa cohort (n = 16) from MD Anderson Cancer Center (Houston, TX, USA). All patients had 3 mpMRI scans during the course of AS. The analysis of our computational technology was performed at both calibration and forecasting stage, considering global metrics (i.e. tumor volume, total tumor cell volume) and local metrics (i.e. Dice score, voxel-wise agreement of tumor cell density between model predictions and mpMRI data). Global results showed an excellent data-model agreement. Local results were promising, but they also suggest model refinements to accommodate local mechanisms of tumor growth. The analysis of personalized calibrations of the model to the 3 mpMRI datasets from each patient at histopathological assessment times (i.e. biopsy, surgery) revealed that the proliferation activity of the tumor and the total tumor cell index are significantly larger in higher-risk PCa cases. We constructed a logistic regression classifier of PCa based on these biomarkers that achieved an excellent classifying performance and could anticipate progression to higher-risk PCa by more than one year.
Thus, while further development and testing in larger cohorts are still needed, these results do suggest that our computational forecasting approach is a promising technology to obtain personalized predictions of PCa growth and progression within the patient’s prostate during AS.
We performed a preliminary study in a small cohort of PCa cases (n = 7) who enrolled in AS at The University of Texas Health Science Center San Antonio (TX, USA), which we then validated in an ensuing study over a larger cohort over a larger PCa cohort (n = 16) from MD Anderson Cancer Center (Houston, TX, USA). All patients had 3 mpMRI scans during the course of AS. The analysis of our computational technology was performed at both calibration and forecasting stage, considering global metrics (i.e. tumor volume, total tumor cell volume) and local metrics (i.e. Dice score, voxel-wise agreement of tumor cell density between model predictions and mpMRI data). Global results showed an excellent data-model agreement. Local results were promising, but they also suggest model refinements to accommodate local mechanisms of tumor growth. The analysis of personalized calibrations of the model to the 3 mpMRI datasets from each patient at histopathological assessment times (i.e. biopsy, surgery) revealed that the proliferation activity of the tumor and the total tumor cell index are significantly larger in higher-risk PCa cases. We constructed a logistic regression classifier of PCa based on these biomarkers that achieved an excellent classifying performance and could anticipate progression to higher-risk PCa by more than one year.
Thus, while further development and testing in larger cohorts are still needed, these results do suggest that our computational forecasting approach is a promising technology to obtain personalized predictions of PCa growth and progression within the patient’s prostate during AS.
The technology developed in this project will contribute to advance the clinical management of PCa from the current observational, population-based standard to a personalized, predictive paradigm. In particular, this technology can bring a major advance in AS protocols by early identifying those patients who require immediate attention and those who can benefit from treatment delay. Our personalized PCa forecasts could guide physicians to personalize and optimize the monitoring strategy, and facilitate the patient-specific estimation of the best timing and type of treatment. Hence, this predictive technology could dramatically decrease the current rates of overtreatment and undertreatment, rationalize the resources devoted to combat PCa, and enhance the quality of life and treatment success of PCa patients. However, further clinical trials will be required before the medical use of the technologies of this project (e.g. involving large patient cohorts and a randomized setting) to test their clinical implementation and prove its benefits. Additionally, this project will contribute to develop the computational modelling of cancer by addressing various challenges in a personalized, predictive clinical setting for the first time: biologically-based inclusion of the main mpMRI data types and clinicopathological variables of PCa in a biomechanistic model, analysis of mechanical effects on both tumor cell mobility and proliferation, and personalized representation of PCa and BPH interactions using a combination of linear and nonlinear biomechanical models. Indeed, this project will produce a unique technology to continue interdisciplinary research in PCa and will facilitate ensuing computational developments. Ultimately, this technology could be combined with a computer-aided diagnosis system and it could also be extended to applications targeting other diseases affecting both men and women, such as other tumors (e.g.: brain, lung, and breast cancers), neurological disorders (e.g.: Alzheimer’s disease, schizophrenia), and cardiovascular conditions (e.g.: atherosclerosis, aneurysms).