Arthur Braat

UMC Utrecht

In short, Arthur Braat's research entails the following:

TowArds IndividuaLized PSMA PET/CT-guided Treatment in Metastastatic PrOstate CanceR Using Machine Learning-Derived Risk Stratification (TAILOR-MADE)

Hormone sensitive metastatic prostate cancer (mHSPC) can present itself in various forms regarding aggressiveness, metastatic burden and prognosis. Accurate disease classification at diagnosis is crucial for accurate treatment selection. Nowadays, PSMA PET/CT is the gold standard, with significantly higher sensitivity to detect metastasis compared to conventional CT and bone scan.Use of PSMA PET/CT instead of bone scintigraphy and CT, results in detection of additional lesions in almost two-thirds of the patients. The current dichotomous classification system, based on the location and number of metastasis diagnosed on bone- and CT scan, classifies patients to “low-volume” [LVD] or “high-volume” [HVD]. Classification to either LVD or HVD has major clinical implications, as different treatment strategies have been shown to be effective in both risk groups.

In today’s clinical routine, CT and bone scanning has been completely replaced by PSMA PET/CT.

As a consequence, the currently available risk stratification system may not be generalizable to the contemporary patient population. In addition, as the mHSPC includes a very heterogeneous patient population, it is attainable that the currently available dichotomous stratification system insufficiently accounts for the heterogeneity in disease prognosis.

Therefore, a new PSMA PET/CT based risk stratification system and contemporary prognostic biomarkers are urgently needed. Not only to improve our current clinical care of men diagnosed with metastatic prostate cancer, but also to ensure the results of future trials remain generalizable to the contemporary state of practice. In addition, availability of a more robust stratification system will improve comparability of in-between trial results and assessment of possible reasons why one trial is positive but the other not.

In this project we will use artificial intelligence (AI) for automatic disease segmentation in PSMA PET/CT’s, image-derived radiomics and identification of predictive features. Together with machine learning and clinical data, we intend to develop a novel AI-based risk classification for mHSPC.

 

Publications

• Integration of prostate-specific membrane antigen-PET and multiparametric MRI for gross tumour volume definition in localised and locally advanced prostate cancer treated with image-guided radiotherapy 
  • Current Opinions in Urology (15-07-2025)
  • Click here to view the publication

 

Short summary of progress/results May 2026

In the second year of this study, the team completed data collection and is nearing completion of the data cleaning. Data cleaning is needed prior to initiating the development of the prediction model, the primary aim of TAILOR-MADE. 

Different segmentation methods and software packages have been evaluated in the second year, upon which the decision was made to develop a new segmentation model for radiomics feature extraction. This will be completed coming year using the available annotated training set.

Short summary of progress/results April 2025

In the first year of this study, the study obtained ethics approval and a dedicated database has been contructed, in collaboration with the TRIPLEAIM-1 investigators.

In light of the multicenter design of the study, majority of procedural and legal documentation has been drafted and approved. A dedicated and secured imaging transfer platform has been initiated. 85% of the anticipated medical centers has approved participation to this multicenter study. We hope to complete data collection the coming months.

The first part of the project will focus on image segmentation on PSAM-PET/CT and extraction of radiomic features. Both in-house developed segmentation model and commercially available segmentation models have been investigated for feasibility.