Category: Clinical Science
At the end of the session, participants will be able to:
- Explain why survival prediction needs refining within WHO grade 3 oligodendroglioma, beyond tumour
grade and molecular markers alone. - Describe how automated measurements of tumour-cell nuclei on routine slides can be linked to patient
survival using a machine-learning model. - Recognize the limits of these image-based risk measures, how few tissue regions are needed, how
region choice affects results, and why independent validation is still required.
COI Disclosure:
None to disclose
Presenter
Adrienn Bourkas – Neuropathology resident at the University of Toronto with an MD and MSc. My research combines diagnostic pathology with machine learning to improve how we predict brain-tumour behaviour from routine slides, with a focus on practical, low-cost tools.
Jinzhen Hu1,2 Kevin Faust2, Parsa Babaei Zadeh2, Adrienn Bourkas1, Shane Eaton1, Andrew Young2, Anzar Alvi2, Nader Alhejaily1, Dimitrios G. Oreopoulos2, Ameesha Paliwal1,2, Assem S. Alrumeh1,3, Andrew Gao1,3, Alberto J. Leon2, Evelyn R. Kamski-Hennekam2, Phedias Diamandis1-4
1 Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, ON, Canada
2 Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada
3 Laboratory Medicine Program, Department of Pathology, University Health Network, Toronto, ON, Canada
4 Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada
Target Audience:
CanMEDS:
Real-time nuclear morphometrics refine survival stratification in WHO grade 3 oligodendroglioma
Abstract
Accurate risk prediction is increasingly important in precision neuro-oncology, particularly for isocitrate dehydrogenase (IDH)-mutant gliomas, where emerging therapies (e.g. vorasidenib) and “watch-and-wait” strategies make it valuable to identify indolent tumours within otherwise high-grade subgroups. Although World Health Organization (WHO) grade stratifies survival in 1p/19q-codeleted oligodendrogliomas (grade 2 versus 3 log-rank p = 0.008, hazard ratio 6.17), grade-3 disease remains clinically heterogeneous, and grade and molecular status alone incompletely predict outcome. We tested whether automated nuclear morphometric features from routine H&E slides could refine survival stratification. Using OnSight Pathology, a free, platform-agnostic tool that applies real-time computer vision to screen captures of any slide viewer, we segmented nuclei (Cellpose-SAM) and extracted morphometric features from 20 tumour patches in each of 64 grade-3 oligodendrogliomas from The Cancer Genome Atlas, then modelled overall survival with an XGBoost survival model under leave-one-out cross-validation. Median nuclear intensity was the dominant predictor (approximately 50% of model gain). Split at its median, this single feature separated the grade-3 cohort into distinct survival groups (log-rank p = 3.1×10⁻⁴, 9/32 versus 1/32 deaths), with lower nuclear intensity marking higher risk (hazard ratio 0.10 per standard deviation, 95% confidence interval 0.01- 0.74, concordance 0.80). In a patch-dilution analysis, a single random patch reproduced significant stratification in 93% of resamples (median log-rank p = 3.7×10⁻³), reaching 100% with three or more patches. We conclude that real-time, hardware-light H&E nuclear morphometrics refine risk stratification within WHO grade 3 oligodendroglioma beyond grade and molecular status alone.