AI-augmented intraoperative decision-making workflows in diffuse midline glioma biopsy using cryosection pathology.

Area of Science:

• Artificial Intelligence in Pathology
• Oncology
• Molecular Diagnostics

Background:

• Intraoperative diagnosis of diffuse midline gliomas via cryosection pathology is crucial but prone to errors, often necessitating re-biopsies.
• Current methods may not fully leverage cryosection images for molecular assessment due to artefacts.

Purpose of the Study:

• To introduce CryoAID, an AI-augmented framework for real-time molecular outcome prediction in diffuse midline glioma during surgery.
• To enhance diagnostic precision and reduce re-biopsy rates by utilizing cryosection images more effectively.

Main Methods:

• CryoAID integrates a generative model for artefact correction and a pathology foundation model for molecular status prediction from cryosection images.
• Validation was performed across internal, multi-centre external, and consecutive patient cohorts.

Main Results:

• CryoAID accurately predicted major molecular statuses (ATRX, H3K27M, TP53) even from images previously unsuitable for molecular examination.
• The AI identified differential clinical features beyond tumor cells, such as glial proliferation and endothelial proliferation.
• Retrospective analyses showed CryoAID reduced re-biopsy rates by 26.4% and 26.6% in internal and consecutive datasets.

Conclusions:

• AI-augmented pathology, exemplified by CryoAID, can extract diagnostic value from cryosection specimens previously deemed non-viable.
• This approach facilitates real-time molecular pathology, potentially minimizing re-biopsies and enhancing diagnostic accuracy for diffuse midline gliomas.