(UroToday.com) The 2024 IBCN annual meeting included a bladder cancer session, featuring a presentation by Dr. David Berman discussing artificial intelligence-enabled bladder cancer grading and external validation of quantitative nuclear features demonstrating better potential for predicting time to recurrence. Bladder cancer grading is crucial for treatment decisions, but the existing ISUP 2004 system is subjective, compromising its reliability and prognostic utility.1
By mathematically defining well-established WHO 2004 grading criteria as quantitative nuclear features and employing artificial intelligence-driven image analysis to extract quantitative nuclear features, Dr. Berman and colleagues previously developed precise, reproducible models for expert consensus grading.2 Quantitative nuclear features also serve as excellent building blocks for prognostic classifiers. At IBCN 2024, using quantitative nuclear features, Dr. Berman and colleagues externally validated grading models and created recurrence-free survival models that outperform grades assigned by pathologists.
All patients in this study were stage Ta, and grading was centrally reviewed as previously described.2 The CNIO Madrid validation cohort comprised 581 0.6mm diameter scanned histopathology images from 403 cases. The Kingston cohort contained 267 1.0mm diameter images from 163 patients. Quantitative nuclear features were extracted using Visiopharm image analysis software. The previously published univariate, decision tree, regression, and random forest models were applied to the CNIO cohort to predict consensus pathologist grade.2 Cox proportional hazards and Random Survival Forest models for time-to-first bladder cancer recurrence were trained and cross-validated in the Kingston cohort using holdout testing.
Random forest grading yielded 80% balanced accuracy in external validation, outperforming all other models:
For survival models, mitotic index, mean lesser diameter (size and shape) and mean variance in hematoxylin intensity (chromatin texture) were most correlated with recurrence-free survival. The Cox proportional hazards model achieved a C-index of 0.73 (95% CI: 0.56-0.88) compared to only 0.55 (95% CI: 0.40-0.69) using consensus grading by 3 genitourinary pathologists. The Random Survival Forest model achieved a C-index of 0.70 (95% CI: 0.66-0.71). Of note, both models dramatically outperformed AUA risk score (0.58, 95% CI: 0.43-0.71):
Dr. Berman concluded his presentation discussing artificial intelligence-enabled bladder cancer grading and external validation of quantitative nuclear features demonstrating better potential for predicting time to recurrence with the following take-home points:
- The external validation results show that models derived from objective and explainable quantitative nuclear features can be reliably extracted and applied to diverse cohorts, highlighting an opportunity to improve the reliability and accuracy of grading
- Quantitative nuclear features can be further leveraged to more accurately identify patients with early and late recurrence, providing an opportunity to identify patients for intensified or relaxed treatment and surveillance.
- These models are currently being adapted to whole slide image analysis.
- Ongoing and future work includes: expanding QNG prognostic classifiers to include stage and progression and WHO Grade 3, externally validating QNG prognostic classifiers on whole slide images, using QNG to improve central review of clinical trials, and working with profession pathology and urology societies to implement validated QNG models in routine clinical practice.
Presented by: David Berman, MD, PhD, Professor, Kingston Health Sciences Centre, Queen’s University, Kingston, Ontario, Canada
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 International Bladder Cancer Network (IBCN) Annual Meeting, Bern, Switzerland, Thurs, Sept 19 – Sat, Sept 21, 2024
References:
- Soukup V, Capounm O, Cohen D, et al. Prognostic performance and reproducibility of the 1973 and 2004/2016 World Health Organization Grading Classification Systems in Non-muscle-invasive bladder cancer: A European Association of Urology Non-muscle invasive bladder cancer Guidelines Panel Systematic Review. Eur Urol. 2017 Nov;72(5):801-813.
- Slotman A, Xu M, Lindale K, et al. Quantitative Nuclear Grading: An Objective, Artificial Intelligence-Facilitated Foundation for Grading Noninvasive Papillary Urothelial Carcinoma. Lab Invest. 2023 Jul;103(7):100155.