Artificial intelligence for streamlining prostate cancer diagnostics

With around 1.2 million cases per year, prostate cancer is the second most common cancer among men. It is usually a slow growing disease that affects older men. It is also a cancer that is heterogenous, often multifocal, and rarely show symptoms as long as it is localized. All these things make the disease difficult to detect, diagnose and study. The objective of this thesis is to develop and improve technologies for prostate cancer diagnostics and to acquire knowledge related to these technologies that directly translate to clinical utility.

In Study I, we extended analysis of the multivariable diagnostic prediction model S3M by exploring the relative contribution from the individual predictors and evaluating the model in reflex setting where the test is only given to men positive on a PSA test. We also updated the list of included predictors and refitted the model to more data. In Study II, we digitized a substantial part of the biopsy cores collected from the men in study I. These images were used to develop and validate an AI for prostate cancer diagnostics by detecting, grading, and measuring the extent of cancer in the biopsies. The AI achieved nearly perfect detection of cancer and expert pathologist level grading of the biopsies. It also well predicted the total tumor burden of the patient. In Study III, we focused our attention on perineural invasion, a common finding in prostate biopsies. This study has added to the evidence that there is substantial and independent prognostic information in this finding and argued that it should be included as a compulsory part in pathology reporting guidelines for prostate biopsies. In Study IV, we developed an AI for detection and localization of perineural invasion in biopsies. The AI achieved high discriminative ability on an independent test set. We are currently collecting external data to validate these results in another environment and to compare the results of the AI against expert pathologists.

In conclusion, the technologies developed in this thesis has shown promise in streamlining the clinical workload around prostate cancer detection and diagnostics. The thesis has also contributed to pieces of information related to these technologies.

List of scientific papers

I. Peter Ström, Tobias Nordström, Markus Aly, Lars Egevad, Henrik Grönberg, and Martin Eklund. The Stockholm-3 Model for Prostate Cancer Detection: Algorithm Update, Biomarker Contribution, and Reflex Test Potential. European Urology. 2018.
https://doi.org/10.1016/j.eururo.2017.12.028

II. Peter Ström*, Kimmo Kartasalo*, Henrik Olsson, Leslie Solorzano, Brett Delahunt, DanielMBerney, David G Bostwick, Andrew J. Evans , David J Grignon, Peter A Humphrey, Kenneth A Iczkowski, James G Kench, Glen Kristiansen, Theodorus H van der Kwast, Katia RM Leite, Jesse K McKenney, Jon Oxley, Chin-Chen Pan, Hemamali Samaratunga, John R Srigley, Hiroyuki Takahashi, Toyonori Tsuzuki, Murali Varma, Ming Zhou, Johan Lindberg, Cecilia Lindskog, Pekka Ruusuvuori, Carolina Wählby, Henrik Grönberg, Mattias Rantalainen, Lars Egevad, and Martin Eklund. Artificial intelligence for diagnosis and grading of prostate cancer in biopsies: a population-based, diagnostic study. LANCET Oncology. 2019. *Equal contribution.
https://doi.org/10.1016/S1470-2045(19)30738-7

III. Peter Ström, Tobias Nordström, Brett Delahunt, Hemamali Samaratunga, Henrik Grönberg, Lars Egevad, and Martin Eklund. Prognostic value of perineural invasion in prostate needle biopsies: a population-based study of patients treated by radical prostatectomy. Journal of Clinical Pathology. 2020.
https://doi.org/10.1136/jclinpath-2019-206300

IV. Peter Ström, Kimmo Kartasalo, Pekka Ruusuvuori, Henrik Grönberg, Hemamali Samaratunga, Brett Delahunt, Toyonori Tsuzuki, Lars Egevad, and Martin Eklund. Detection of Perineural Invasion in Prostate Needle Biopsies with Deep Neural Networks. [Manuscript]