Deep Learning-Based Treatment Planning for PSMA RPT

Project Summary Prostate cancer (PCa) is one of the most prevalent cancers worldwide, with an estimated 1,111,700 new cases and 307,500 deaths in 2012, and is currently the second most common cancer in men, and fifth leading cancer mortality in men. Prostate-specific membrane antigen (PSMA) is a transmembrane marker which shows limited expression levels in healthy tissues such as colon and proximal tubules in kidneys, but is recognized as the most prostate-specific antigen yet identified. This has led to the rapid expansion of radiopharmaceutical therapy (RPT) in the treatment of metastatic castration resistant prostate cancers (mCRPC) using 177Lu-PSMA compounds. Subjects receiving this treatment undergo pre-therapy diagnostic PSMA-PET scans, and can undergo SPECT/CT scans after each round of therapy due to the theranostic nature of 177Lu. This marks a major advantage of RPT over chemotherapies, as the biodistribution of the therapeutic agent can be visualized with PET or SPECT – which enables the collection of pharmacokinetic data to perform dosimetry calculations to optimize treatment planning. Unfortunately, to date, all prostate RPT use a standard dose regimen of 6x200mCi irrespective of the size of the patient or the pharmacokinetics and biodistribution of the radiopharmaceutical which govern the actual dose imparted to both the tumor and at-risk healthy tissues. Here, we propose to address this limitation by utilizing the wealth of imaging data that can be generated in the course of 177Lu-PSMA therapy. We propose to build models of 177Lu-PSMA dosimetry using imaging information available from pre-treatment 68Ga- or 18F-PSMA PET as well as 177Lu-PSMA SPECT obtained during RPT to personalize the dose injected at each cycle. We propose to develop deep learning-assisted predictive models that allow for personalized estimation of dosimetry, and thus personalized treatment planning. Clinical markers such as PSA levels and GFR of ClCr will be included as predictors.