Artificial Intelligence and Deep Learning in Prostate Cancer Pathology

(UroToday.com) The 2024 American Society of Clinical Oncology (ASCO) annual meeting featured a session on applications of artificial intelligence in prostate cancer care, and a presentation by Dr. Liang Cheng discussing artificial intelligence and deep learning in prostate cancer pathology. Dr. Cheng started this presentation by noting that the recently published Lancet Commission on Prostate Cancer highlights several sobering statistics:¹

• The number of new cases will rise from 1.4 million in 2020 to 2.9 million in 2040
• The projected rise cannot be prevented by lifestyle changes or public health interventions
• Late diagnosis of prostate cancer is widespread worldwide
• Early diagnosis systems need to integrate the growing power of artificial intelligence to aid interpretation of biopsy specimens

With regards to the pathologist workforce, the total number of pathologists has decreased in each year (2007 to 2017) in the US and Canada, for an overall decrease of 17.5% (15,568 to 12,839), with the diagnostic workload increasing by 41.7%. This is in the face of the total number of physicians increasing by 16.6% during that same time frame (765,688 to 892,856). Perhaps artificial intelligence is the solution?

There are many aspects of artificial intelligence in prostate cancer pathology, including (i) cancer detection/diagnosis, (ii) Gleason grading, (iii) molecular classification, and (iv) outcome prediction. A recent systematic review and meta-analysis of artificial intelligence diagnostic accuracy of prostate cancer histology identification and grading noted an impressive AUC of 0.9885.² Artificial intelligence across all spectrums of digital pathology detection/diagnosis notes that for prostate cancer, sensitivity is 95% and specificity is 96%. However, there are several areas of concern, including risk of bias, patient selection, the index test, and the reference standard.

In a study assessing artificial intelligence modeling for detecting pathological lymph node metastasis, the following workflow was utilized:³
Among 1,297 patients (8,225 slides), the AUC for lymph node detection was 0.989. Moreover, this artificial intelligence algorithm boosted the diagnostic sensitivity of two junior pathologists (both p = 0.041) to the level of senior pathologists (both p > 0.99) and substantially reduced the four pathologists’ slide reviewing time.

For Gleason grading, Dr. Cheng notes that the 2019 ISUP modified Gleason Scores were recently developed into prognostic Gleason Grade Groups (ISUP/WHO Grades):

In 2020, Steiner and colleagues evaluated an expert-level artificial intelligence-based assistive tool when used by pathologists for the grading of prostate biopsies.⁴ Among biopsies from 240 patients with a median PSA level of 6.5 ng/mL, artificial intelligence-assisted review by pathologists was associated with a 5.6% increase in agreement with subspecialists across all biopsies and a 6.2% increase in agreement with subspecialists for grade group 1 biopsies:
Additionally, artificial intelligence assistance was also associated with improvements in tumor detection, mean review time, mean self-reported confidence and inter-pathologist agreement.

Dr. Cheng then discussed molecular classification and artificial intelligence. One example is that attention-based deep learning has demonstrated the feasibility of being able to predict TMPRSS2:ERG fusion status. In a recent study of artificial intelligence-assisted genome interpretation, Woodcock and colleagues investigated the genomic evolution of prostate cancer through the application of three separate classification methods, each designed to investigate a different aspect of tumor evolution.⁵ They found that integrating the results revealed two distinct types of prostate cancer that arise from divergent evolutionary trajectories:
Hiremath et al. recently evaluated the added benefit of integrating features from pre-treatment MRI (radiomics) and digitized post-surgical pathology slides (patronymics) in prostate cancer patients for prognosticating outcomes post radical-prostatectomy.⁶ Among 58 prostate cancer patients, all underwent pre-treatment 3-T MRI before radical prostatectomy and radiomic and pathogenic features were extracted from prostate cancer regions on MRI and prostatectomy specimens:
Overall, the integrated radio-pathomic model M (AUC = 0.80) outperformed the radiomic (AUC = 0.57) and pathomic (AUC = 0.76) models alone in predicting extracapsular extension.

Dr. Cheng then highlighted several challenges in clinical implementation of artificial intelligence in pathology:

• Difficulty in generalization and explainability
• Lack of high-quality open source datasets
• Difficulty in tracking rare conditions and ethnic variability
• Clinical workflow integration
• Regulatory barrier and reimbursement issues

What may also be on the horizon is whether ChatGPT can replace a pathologist. The following question was posed to ChatGPT: Can you explain what this image shows (biopsy specimen)? Answer: The changes are consistent with prostate cancer. If you have concerns about the diagnosis, it is always best to discuss with a healthcare professional.

 Dr. Cheng concluded his presentation discussing artificial intelligence and deep learning in prostate cancer pathology with a quote from Dr. William B. Schwartz from the New England Journal of Medicine in 1970, which still rings true regarding artificial intelligence in 2024: “Because the coming intervention of the information sciences represents radical change, it will also inevitably create radical problems. The power of the information sciences is such that it will alter the face of medicine, and we can ill afford to ignore this impending reality. Physicians must take the opportunity to assist and lead in the planning and implementation of a system that can best serve the interests of both the public and the medical community.”

Written by: Zachary Klaassen, MD, MSc – Urologic Oncologist, Associate Professor of Urology, Georgia Cancer Center, Wellstar MCG Health, @zklaassen_md on Twitter during the 2024 American Society of Clinical Oncology (ASCO) Annual Meeting, Chicago, IL, Fri, May 31 – Tues, June 4, 2024.

References:

1. James ND, Tannock I, N’Dow J, et al. The Lancet Commission on prostate cancer: Planning for the surge in cases. Lancet. 2024 Apr 27;403(10437):1683-1722.
2. Morozov A, Taratkin M, Bazarkin A, et al. A systematic review and meta-analysis of artificial intelligence diagnostic accuracy of prostate cancer histology identification and grading. Prostate Cancer Prostatic Dis. 2023;26(4):681-692.
3. Wu S, Wang Y, Hong G, et al. An artificial intelligence model for detecting pathological lymph node metastasis in prostate cancer using whole slide images: A retrospective, multicentre, diagnostic study. EClinicalMedicine. 2024 Apr 5;71:102580.
4. Steiner DF, Nagpal K, Sayres R, et al. Evaluation of the use of combined artificial intelligence and pathologist assessment to review and grade prostate biopsies. JAMA Netw Open. 2020 Nov 2;3(11):e2023267.
5. Woodcock DJ, Sahil A, Teslo R, et al. Genomic evolution shapes prostate cancer disease type.  Cell Genomics 2024. Mar 13;4(3):100511.
6. Hiremath A, Corredor G, Li L, et al. An integrated radiology-pathology machine learning classifier for outcome prediction following radical prostatectomy: Preliminary findings. Heliyon 2024 Apr 15;10(8):e29602