Ultra-fast, deep-learned CNS tumour classification during surgery

Abstract

Primary treatment of central nervous system (CNS) tumours includes neurosurgical resection, in which a delicate balance must be struck between maximizing extent of resection and minimizing the risk of neurological damage. However, neurosurgeons have limited knowledge of the precise tumour type during surgery. Current standard practice relies on preoperative imaging and intraoperative histological analysis, but these are not always conclusive and occasionally wrong. Using rapid nanopore sequencing, a sparse methylation profile can be obtained during surgery. We developed Sturgeon — a patient-agnostic transfer-learned neural network, to enable ultra-fast molecular (sub)classification of CNS tumours based on such sparse profiles. In the vast majority of the >50 cases intra-operatively analysed so far, Sturgeon delivered an accurate diagnosis within 40–60 minutes after starting sequencing; in almost all remaining cases the required confidence threshold was not reached. We conclude that machine-learned diagnosis based on low-cost intraoperative sequencing can assist intra-operative neurosurgical decision making in a highly clinically relevant way.

Biography

Pieter Wesseling, MD, PhD, is a clinical (neuro)pathologist and full professor in neuro-oncological pathology, affiliated with Amsterdam University Medical Centers and the Princess Máxima Center for Pediatric Oncology, Utrecht, Netherlands. Pieter has led multiple national and international research projects, and coauthored >300 papers on neuro-oncological topics in international journals. He has also served as an expert-editor of the 2021 edition of the WHO CNS tumor classification and is chair of the consortium to Improve Molecular and Practical Approaches for CNS tumor Taxonomy (cIMPACT-NOW).