Paul van Diest

UMC Utrecht

In short, Paul van Diest's research entails the following: 

Cost savings in pathology by artificial intelligence: prospective clinical trials paving the way for implementation of artificial intelligence  

'AI has shown great promise in improving speed and quality of pathology diagnostics. However, implementation of AI is yet lacking because of a problematic business case: license costs for AI algorithms on top of an expensive fully digital diagnostics workflow cannot easily be paid, requiring tangible cost savings by AI.

In this project, we run two prospective clinical trials in daily pathology practice that will, for the first time, make clear to which extent two commercially available clinical grade AI algorithms for detection of prostate cancer and metastases in breast cancer sentinel lymph nodes lead to tangible cost savings by obviating immunohistochemistry. In the control arm, cases will be reviewed as usual, applying immunohistochemistry-or-not at the discretion of the pathologist. In the intervention arm, pathologists will assess cases after the algorithm has processed these. We expect pathologists to need less immunohistochemistry to detect/reliably measure tumor load of prostate cancer and find SN metastases in breast cancer.

With the expected reduction in immunohistochemistry, we hope to show that AI implementation and running costs can be earned back in 1-2 years. These trials will help to build the business case for implementation of these AI algorithms in clinical pathology practice, and thereby close the translational gap.' 

 

Publications

AI helpt bij opsporen uitzaaiingen borstkanker - UMC Utrecht (Dutch only)

Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: the CONFIDENT-B single-center, non-randomized clinical trial

 

Final results November 2025

The aim of this project was to evaluate in two prospective clinical trials in prostate and breast cancer to which extent two commercially available AI algorithms lead to tangible cost savings by obviating immunohistochemistry (IHC). This would help to build the business case for implementation of these AI algorithms in pathology practice, closing the translational gap.

At the department of pathology of the UMC Utrecht, we implemented two CD-IVDR/FDA approved AI tumor detection algorithms in a prospective clinical trial setting in daily practice in prostate cancer (CONFIDENT-P) and breast cancer sentinel nodes (CONFIDENT-B), enabling comparison between AI- and non-AI assisted workflows. Patient safety was not be compromised, as output of algorithms was pathologist supervised, and IHC was performed in all cases deemed negative by AI-assisted pathologists 1 .

In the CONFIDENT-B trial 2,3 , (N=190), the Visiopharm algorithm detected all macro- and micrometastases and 44% of isolated tumor cells. There was significantly less (p<0.02) use of IHC in the AI supported arm (RR: 0.680, 95%CI: 0.347-0.878). Potential IHC cost savings at UMCU were estimated at 40k euros annually. Time spent was 40% less in the AI supported arm. All pathologists were happy to use the algorithm. The algorithm is now used routinely at UMCU, running in the background for all breast cases with lymph nodes.

For the CONFIDENT-P trial 4,5 (N=82) the Paige algorithm for prostate cancer detection was cloud implemented. AI assistance significantly reduced the risk of IHC use per detected PCa case at both the patient level (RR, 0.55; 95% CI, 0.39 to 0.72) and slide level (RR, 0.41; 95% CI, 0.29 to 0.52). Cost reductions on IHC were €1,700 for the trial, at €50 per IHC stain. AI-assisted pathologists reported higher confidence in their diagnoses (80% vs. 56% confident or high confidence). The median assessment time per HE slide showed no significant difference between the AI-assisted and control arms (139 seconds vs. 112 seconds; P = 0.2).

In conclusion, we have successfully shown in two unique prospective clinical trials (breast cancer lymph nodes and prostate cancer detection) that a significant cost reduction can be achieved with AI implementation by reducing IHC. Both algorithms have been implemented routinely at the UMC Utrecht, which means that also the pathology lab at the Gelre hospital (and soon the Meander hospital) will profit from these implementations.

1. Flach RN, Stathonikos N, Nguyen TQ, Ter Hoeve ND, van Diest PJ, van Dooijeweert C. CONFIDENT-trial protocol: a pragmatic template for clinical implementation of artificial intelligence assistance in pathology. BMJ Open 2023 Jun 7;13(6):e067437.
2. van Dooijeweert C, Flach RN, Ter Hoeve ND, Vreuls CPH, Goldschmeding R, Freund JE, Pham P, Nguyen TQ, van der Wall E, Frederix GWJ, Stathonikos N, van Diest PJ. Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: the CONFIDENT-B single-center, non-randomized clinical trial. Nat Cancer 2024;5:1195-1205.
3. van Dooijeweert C, Flach RN, Ter Hoeve ND, Vreuls CPH, Goldschmeding R, Freund JE, Pham P, Nguyen TQ, van der Wall E, Frederix GWJ, Stathonikos N, van Diest PJ. Author Correction: Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: the CONFIDENT-B single-center, non-randomized clinical trial. Nat Cancer 2024;5:1285. Erratum for: Nat Cancer. 2024 Aug;5(8):1195-1205.
4. Flach RN, van Dooijeweert C, Nguyen TQ, Lynch M, Jonges TN, Meijer RP, Suelmann BBM, Willemse PM, Stathonikos N, van Diest PJ. Prospective Clinical Implementation of Paige Prostate Detect Artificial Intelligence Assistance in the Detection of Prostate Cancer in Prostate Biopsies: CONFIDENT P Trial Implementation of Artificial Intelligence Assistance in Prostate Cancer Detection. JCO Clin Cancer Inform 2025;9:e2400193.
5. PhD thesis “Improving prostate cancer detection and grading using novel digital techniques” by Rachel N Flach. Utrecht University, October 2, 2024.

 

Brief summary of progress / results November 2023

Half way the project is running ahead of schedule. The CONFIDENT-B trial has been very successfully completed, and showed that up to 40k euros can be saved annually on immunohistochemistry of sentinel nodes at the UMC Utrecht. This has already led to full integration into the digital diagnostics workflow of the department of pathology of the UMCU, and routine use has just started.  The patient accrual for the CONFIDENT-P trial has already been completed. We expect to finish data analysis and manuscript writing in early 2024. This means that we will have time for one or more CONFIDENT trials for other organs such as testing the Visiopharm lymph node metastases app for melanoma, or lung or cervical or head&neck cancer.