(MRC DTP) Using advanced MR imaging to select the best treatment for men with high risk prostate cancer

High-risk prostate cancer patients receive three months of androgen deprivation therapy (ADT) followed by 20 sessions of radiotherapy. Currently, the risk stratification for these patients is based on presenting parameters, with limited information about disease spread within the prostate. The introduction of multi-parametric magnetic resonance imaging (mpMRI) into the diagnostic pathway has improved the sensitivity and specificity of detecting prostate cancer, and therefore informing the most appropriate treatment for each man [1]. However, this information does not inform how an individual’s prostate cancer will respond to treatment. It is the case that we currently over-treat men with less aggressive disease and under-treat men with more aggressive disease. Studies with mpMRI have shown changes during treatment [2-3], however, these studies are based on a limited set of imaging timepoints, and high-risk patients would benefit from a greater understanding of their disease.

This project will collect a uniquely detailed dataset with mpMRI at each radiotherapy session, utilising the MRI-linac, for 50 patients with high-risk prostate cancer. The MRI-linac allows the mpMRI to be acquired during standard treatment enabling this data collection. mpMRI will be designed to provide comprehensive information of a patient’s disease. Due to ADT and radiotherapy homogenising the signal in the prostate we will use a radiomics feature analysis approach. This approach has shown the feasibility of radiomics and mpMRI to distinguish cancerous and healthy tissue in the prostate [4]. This high temporal resolution dataset will be used to investigate changes in radiomic features during treatment. Radiomics extracts details information directly from patient images and is based on histogram, textural and filter-based techniques[5]. These features can be extracted across the mpMRI and changes across treatment will be quantified for each patient.

Changes in image features, with patient demographics, will be used to train a supervised machine learning algorithm. This approach will derive a radiomics signature, as a function of features seen pre-ADT or pre-radiotherapy. Therefore, allowing our approach to be applicable for all men with high-risk prostate cancer as part of their standard diagnostic pathway. This signature will be designed to predict image feature changes and, particularly, convergence of features from cancerous and healthy prostate tissue. Validation will be performed on an independent dataset of 100 patients, with mpMRI pre-ADT and/or pre-radiotherapy and with sufficient follow-up for biochemical recurrence data to be available, 3-5 years. Sensitivity and specificity of our approach will be calculated.

https://www.research.manchester.ac.uk/portal/en/researchers/marcel-van-herk(80d5752d-51ed-421e-ab67-8eff5f6becc9).html