Derya Yakar

UMCG

In short, Derya Yakar's research entails the following:

Artificial Intelligence-based MRI diagnosis of Prostate Cancer: a two-step research approach to realize clinical implementation

This research project is dedicated to developing and implementing a (semi-) autonomous AI system for MRI reporting in the diagnosis of clinically significant prostate cancer (csPCa). The project is aimed at overcoming current challenges such as the shortage of expert prostate radiologists and the increasing demand for MRI scans. The PI-CAI algorithm, originally developed using data from over 12,000 prostate MRI cases, will be elevated to higher levels of automation. This enhancement is expected to significantly improve workflow efficiency and address the capacity challenges in radiology and healthcare.

The project is structured in two primary steps. The first step includes meeting four essential preconditions: ensuring patient acceptance of AI, developing safety control measures, generating prospective pilot data to evaluate safety, diagnostic performance, and workflow efficiency, and preparing for a Randomized Controlled Trial (RCT). The second step involves initiating and conducting the RCT to rigorously test and validate the AI system's effectiveness and practicality in a clinical setting.
Collaborators include co-directors Thomas Kwee (UMCG) and Henkjan Huisman (Radboudumc). Marieke van Gerner-Haan (RUG), Yfke Ongena (RUG), Igle Jan de Jong (UMCG), Thea van Asselt, Dennis Rouw (Martini Hospital), and Maarten de Rooij (Radboudumc).

 

Publications

• Patient perspectives on the use of artificial intelligence in prostate cancer diagnosis on MRI
  • PubMed, August 14, 2024
  • Click here to view the publication

 

Brief summary of progress / results July 2026

This project aims to develop and implement semi-autonomous AI for diagnosis of clinically significant prostate cancer (csPCa) in MRI in two phases.  phase one started in September 2024 and focuses on fulfilling the four preconditions. The current studies provide the evidence base necessary for a prospective pilot study that compares semi-autonomous AI with radiologists.

In a narrative review, the remaining technical and societal barriers to the clinical implementation of autonomous AI-based csPCa detection were explored. Particularly, the review focused on the current evidence base for autonomous csPCa detection, safety issues and mitigation strategies, as well as perspectives of patients and radiologists. The findings suggest a need for multicenter prospective trials, further validation of safety mechanisms, as well as transparency and education about AI for patients and radiologists. The manuscript has been submitted to a scientific journal and is undergoing peer review.

A multicenter patient survey is being developed to investigate patients’ willingness to participate in a randomized controlled trial comparing uncertainty-based, semi-autonomous AI with radiologist assessment. Additionally, an animated video will be used to examine whether clear information about the AI system and its safeguards influences patients’ willingness to participate.

A multicenter reader study is currently being held on the Grand Challenge platform among 46 radiologists worldwide. The reader study aims to validate the case-level diagnostic performance and workload reduction of an uncertainty-based semi-autonomous workflow. The study is expected to provide evidence for the design of a future prospective pilot study.

The lesion-level performance of uncertainty-based semi-autonomous AI was assessed in the context of MRI-guided biopsies. Additionally, two new lesion-level metrics were introduced. The analyses have been completed and suggest semi-autonomous AI may help to reduce unnecessary biopsies in healthy patients, while radiologist supervision may still be needed for high-certainty csPCa exams. The manuscript is being prepared for submission to a scientific journal.

Brief summary of progress / results July 2025

This project aims to develop and implement semi-autonomous AI for diagnosis of clinically significant prostate cancer (csPCa) in MRI in two phases. Phase  one has started in September 2024 and focuses on fulfilling the four preconditions.

Two studies are being conducted to validate the semi-autonomous AI system and investigate its safety. For this, the pipeline to generate outcomes with a developed AI model (PI-CAI) and to calibrate the uncertainty thresholds have been implemented. MRI data have been prepared from two Dutch medical centers, and additional data is being added.

The first study is a multi-center reader study that aims to validate the case-level diagnostic performance and workload reduction of the semi-autonomous workflow. The study design has been finalized, and a power analysis has been conducted. The reader study has been set up on Grand Challenge and will start in the third quarter of 2025.

The second study assesses the lesion-level performance of certain MRI exams within the same semiautonomous workflow. We introduce two metrics to measure diagnostic performance in the context of guided biopsies. Results suggest potential to reduce unnecessary biopsies in non-csPCa exams and highlight the need for future research on reaching expert-level performance and determining appropriate lesion-level metrics.