Related Experiment Video
Updated: Jun 23, 2025
09:53
Role of Diffusion MRI Tractography in Endoscopic Endonasal Skull Base Surgery
Published on: July 5, 2021
3.6K
A Predictive Model for Intraoperative Cerebrospinal Fluid Leak During Endonasal Pituitary Adenoma Resection Using a
Faraz Behzadi1, Mohammad Alhusseini2, Seunghyuk D Yang1
1Department of Neurological Surgery, Loyola University Medical Center, Maywood, Illinois, USA.
World Neurosurgery
|June 14, 2024
Summary
Identifying cerebrospinal fluid (CSF) leaks during surgery is crucial. A machine learning model accurately predicted intraoperative CSF leaks using MRI, highlighting specific anatomical areas and visual disturbances as key risk factors.
Area of Science:
- Neurosurgery
- Medical Imaging
- Artificial Intelligence
Background:
- Cerebrospinal fluid (CSF) leaks during endoscopic endonasal transsphenoidal surgery pose a risk for postoperative complications.
- Identifying clinical and anatomical risk factors for intraoperative CSF leaks is essential but not well-defined.
- Preoperative magnetic resonance imaging (MRI) data can be leveraged to predict these risks.
Purpose of the Study:
- To identify clinical and anatomical risk factors for intraoperative cerebrospinal fluid (CSF) leaks.
- To develop and validate a machine learning model for predicting intraoperative CSF leaks using preoperative MRI.
- To enhance surgical planning and patient safety by anticipating potential CSF leaks.
Main Methods:
- A retrospective analysis of adult patients undergoing endoscopic endonasal transsphenoidal surgery with accessible preoperative stereotactic MRI.
- Application of a two-dimensional (2D) convolutional neural network (CNN) model trained on stereotactic T2-weighted brain MRI scans.
- Statistical analysis to identify demographic, clinical, and anatomical risk factors associated with intraoperative CSF leak.
Main Results:
- Out of 220 patients, 81 (36.8%) experienced intraoperative CSF leak.
- Visual disturbance was the only statistically significant clinical risk factor identified (P=0.008).
- The 2D CNN model achieved 92% accuracy, with an AUC of 0.90, in predicting CSF leaks, with class activation mapping highlighting CSF flow regions as critical predictors.
Conclusions:
- The 2D CNN model effectively predicts intraoperative CSF leaks with high accuracy.
- Anatomical regions including the diaphragma sellae, clinoid processes, temporal horns, and optic nerves are correlated with CSF leak risk.
- The model's findings, combined with the identification of visual disturbances as a clinical risk factor, can aid surgical teams in anticipating and preparing for intraoperative CSF leaks.
