(UroToday.com) The 2024 IBCN annual meeting included a session on treatment response correlates, featuring a presentation by Dr. Jacqueline Fontugne discussing the quantification of intra-tumoral molecular subtype heterogeneity in muscle invasive bladder cancer from histological slides using a deep learning approach in the VESPER trial.1,2 Dr. Fontugne and colleagues have previously demonstrated in the VESPER trial that muscle invasive bladder cancer patients with molecular subtype intra-tumoral heterogeneity show poor prognosis in a cohort of chemotherapy-treated patients:
The goal of this study was to determine whether a deep learning-based approach applied to histological slides can identify molecular subtypes, improve subtype heterogeneity quantification, and its association with outcome:
The investigators obtained consensus molecular subtypes from one or multiple tumor areas from TURBT specimens from the VESPER trial (n = 417 selected areas from 300 muscle invasive bladder cancers). They then constructed a deep-learning model to predict subtypes at a tile level (400 x 400μm) through training on the 417 areas from whole-slide HES-stained diagnostic TURBT images. The model was applied at the whole slide level to construct subtype heterogeneity maps:
The model will be validated in external muscle invasive bladder cancer cohorts (TCGA-BLCA, COBLAnCE NAC).
This model predicted consensus molecular subtypes with a receiver operating characteristic AUC of 0.8750 and an accuracy of 0.6121. Visualization of the most predictive tiles for each subtype suggested some associations with known morphological features:
Subtype maps showed a variety of heterogeneity profiles, which were quantified as the percentage of tumor tiles assigned to a given subtype. Preliminary performance metrics when applying this model to the COBLAnCE cohort showed a receiver operating characteristic AUC of 0.9121 and an accuracy of 0.708.
Dr. Fontugne concluded her presentation discussing the quantification of intra-tumoral molecular subtype heterogeneity in muscle invasive bladder cancer from histological slides using a deep learning approach in the VESPER trial with the following take-home points:
- Consensus molecular subtypes can be predicted using artificial intelligence on histopathology slides
- Model improvements are ongoing with refinement to detect “basalness”
- Validation in external cohorts and through spatial transcriptomics are ongoing
- Direct artificial intelligence prediction of outcomes is possible from H&E slides
Presented by: Jacqueline Fontugne, MD, Institut Curie, Saint-Cloud, France
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:- Pfister C, Gravis G, Flechon A, et al. Dose-dense methotrexate, vinblastine, doxorubicin, and cisplatin or gemcitabine and cisplatin as perioperative chemotherapy for patients with nonmetastatic muscle-invasive bladder cancer: Results of the GETUG-AFU V05 VESPER trial. J Clin Oncol. 2022 Jun 20;40(18):2013-2022.
- Pfister C, Gravis G, Flechon A, et al. Perioperative dose-dense methotrexate, vinblastine, doxorubicin, and cisplatin in muscle-invasive bladder cancer (VESPER): Survival endpoints at 5 years in an open-label, randomized, phase 3 trial. Lancet Oncol. 2024 Feb;25(2):255-264.