75. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie

Professer

Vorsitz

Prof. Dr. med.

Dr. med.

Prof. Dr.

Predictability of lesion locations in left-sided brain tumor patients with impaired proper name retrieval

Tumor

Prädiktivität von Läsionslokalisationen bei Patient*innen mit Benennstörung bekannter Personen und Orte und linksseitigen Hirntumoren

Abstractvortrag | Abstract talk

PD Dr.

Streamlined intraoperative Stereotactic-guided biopsy analysis: Brain tumor detection, classification and molecular subtyping with stimulated raman histology and deep learning

Prozessoptimierte intraoperative Gewebeanalyse von stereotaktischen Biopsien: Detektion, Klassifikation und molekulare Subtypisierung von Hirntumoren mit stimulierter Raman-Histologie und Deep Learning

Deep-learning-based algorithms leveraged intraoperative decision-making via stimulated Raman histology (SRH) during microsurgical resection. This study evaluates this technique for fast diagnosis during stereotactic-guided brain biopsies and defines a minimum tissue sample size, ensuring diagnostic accuracy.A prospective single-center study evaluated 121 SRH images from 84 patients with suspected intracranial lesions undergoing stereotactic-guided biopsy. Fresh, label-free tissue samples were squeezed on a slide and imaged with a portable fiber-laser Raman scattering microscope. Deep-learning-based models were tested to (1) identify tumorous/non-tumorous tissue for qualitative biopsy control, then (2) classify it into malignant glioma, diffuse low-grade glioma, metastases, lymphoma, or gliosis, and then (3) molecularly subtype the IDH- and 1p/19q-status of adult-type diffuse gliomas. Model predictions were compared to frozen section analyses and final neuropathological diagnoses.The first model identified tumorous/non-tumorous tissue with 91.5% accuracy. Tissue sample size influenced brain tumor subclassification accuracy (81.6%, κ=0.72 frozen section; 73.9%, κ=0.61 second model). SRH slide images &gt;140 high-quality patches with a mean squeezed sample of 5.26mm2 yielded 89.5% accuracy in brain tumor subclassification comparable to conventional frozen section diagnostics and 93.9% accuracy in molecular subtyping of adult-type diffuse gliomas outperforming the limits of conventional rapid section histopathological review. Of note, the tissue sample size subject to SRH was still significantly smaller (4.1±2.5mm2) than H&amp;E frozen sections (16.7±8.2mm2, p&lt;0.001).AI-based image analysis is non-inferior to frozen section analysis in detecting and classifying brain tumors during small stereotactic-guided biopsies once a critically squeezed sample size is reached. It also allows for valid molecular glioma subtyping, potentially leading to faster treatment decisions in the future. Further refinement is needed for long-term clinical routine use.

Univ.-Prof. Dr. med.

Structural connectivity alterations in patients with tumor lesions in the supplementary motor area (SMA)   

Veränderungen der strukturellen Konnektivität bei Patienten mit Tumoren in der supplementär-motorischen Area (SMA)

Prof

Preservation of language during awake craniotomy for infiltrating brain tumors – A new diagnostic tool improves functional resolution

Optimierte Sprachtestung bei Wach-Operationen von infiltrierenden Hirntumoren

PD. Dr.

Voxel-based analysis of the hemorrhage risk after stereotactic and robotic brain biopsies

Voxel-basierte Analyse des Blutungsrisikos nach stereotaktischen und robotergestützten Hirnbiopsien

Entwicklung und Validierung eines deep learning basierten Modells zur präoperativen Unterscheidung von Glioblastomen und zerebralen Lymphomen vor stereotaktischer Biopsie

Development and validation of a deep learning-based model for the preoperative differentiation of glioblastomas and cerebral lymphomas prior to stereotactic biopsy

Session 22
Tumor – Verschiedenes 2

Appointment

Vorsitz

Topic

Programme

Session 22Tumor – Verschiedenes 2

Appointment

Vorsitz

Topic

Programme

Session 22
Tumor – Verschiedenes 2