Postoperative timing effects on AI-based resection-cavity segmentation in glioma: multi-site evaluation and relevance assessment
Abstract
Recent evidence underscores the need for early post-surgical cavity delineation to improve brain tumor treatment. Manual segmentation—the gold standard—is slow and labor-intensive but necessary to link early post-surgical lesions to brain connectivity and functional outcome. Using one of the largest post-operative glioma MRI datasets available to date ( N = 1056), (i) we evaluated the effect of postoperative scan timing on AI-based resection-cavity segmentation; (ii) we investigated whether automated segmentations replicate findings showing that surgery-induced network disconnection severity predicts worse language fluency scores. While the segmentation of chronic cavities yielded accurate results, early postoperative segmentations initially posed challenges due to domain shift. Semi-supervised learning and model fine-tuning overcame this issue, providing preliminary evidence supporting the adoption of AI-driven disconnectomics. We provide a robust AI-driven tool for early postsurgical assessment and functional recovery prediction in neuro-oncological care, showing that the time between surgery and cavity delineation is critical for AI reliability.
// Source
Authors: Ludovico Coletta, Alberto Spolverato, Luca Zigiotto, Riho Nakajima, Mitsutoshi Nakada, Alberto Morello, Elena Garro, Fabio Cofano, Diego Garbossa, Naomi Houedjissin, Christian Rubbert, Michael S. Sabel, Marion Rapp, Iñigo Pomposo, Manuel Carreiras, Laura de Frutos-Sagastuy, Lucía Amoruso, Hugues Duffau, Paolo Avesani, Silvio Sarubbo
Institutions: Düsseldorf University Hospital, Heinrich Heine University Düsseldorf, University of the Basque Country, Inserm, Centre National de la Recherche Scientifique, Kanazawa University, University of Turin, Centre Hospitalier Universitaire de Montpellier, University of Trento, Université de Montpellier, Ikerbasque, BioCruces Health research Institute, Ospedale Santa Chiara, Hôpital Gui de Chauliac, Hospital de Cruces, Institut de Génomique Fonctionnelle, Fondazione Bruno Kessler, Basque Center on Cognition, Brain and Language