PhD thesis

Artificial intelligence as a digital fellow in pathology

Human-machine synergy for improved prostate cancer diagnosis

I defended this at Radboud University in 2022. This is the web edition, one page per chapter, with the figures and the references in place.

Download the PDF Repository record

The printed version, with the full reference list, is in the Radboud Repository.

Contents

1. Ch. 1

   General introduction

   How prostate cancer is diagnosed and graded, why Gleason grading varies so much between pathologists, and what deep learning brings to the problem.

2. Ch. 2

   IHC-based epithelium segmentation

   Restaining slides with immunohistochemistry gave us a precise ground truth. One network learned epithelium on IHC, a second learned it on H&E.

   Scientific Reports (2019) · doi:10.1038/s41598-018-37257-4

3. Ch. 3

   Automated Gleason grading of prostate biopsies

   We built a deep-learning system that grades prostate biopsies without manual annotations. It matched the expert consensus and outperformed 10 of 15 pathologists.

   The Lancet Oncology (2020) · doi:10.1016/S1470-2045(19)30739-9

4. Ch. 4

   The PANDA Challenge

   A global competition on 10,616 public biopsies drew 1,290 developers. The resulting algorithms matched pathologist-level grading on blinded US and EU cohorts.

   Nature Medicine (2022) · doi:10.1038/s41591-021-01620-2

5. Ch. 5

   AI-assisted Gleason grading

   Fourteen pathologists graded 160 biopsies with and without AI assistance. With the AI they agreed more with the expert reference, and beat the standalone system.

   Modern Pathology (2021) · doi:10.1038/s41379-020-0640-y

6. Ch. 6

   General discussion

   A look back at the four studies and forward: how AI support fits a pathologist's workflow, and what diagnostics could look like beyond Gleason grading.

7. Summary

   Summary

   The thesis in one page: AI trained to grade prostate biopsies reaches pathologist level, and improves further once pathologists use it themselves.

8. Samenvatting

   Samenvatting

   Het proefschrift in het kort: AI die prostaatbiopten beoordeelt haalt het niveau van pathologen, en werkt nog beter zodra pathologen het zelf gebruiken.

References are numbered per chapter here, so the numbers do not match the printed thesis.