Predicting gross-total resection in skull base chondrosarcoma: a multi-institutional machine learning study
Juan Pablo Zuluaga-Garcia1, Franco Rubino2, Esteban Ramirez Ferrer1
11Department of Neurosurgery, The University of Texas MD Anderson Cancer Center, Houston, Texas.
Journal of Neurosurgery
|July 24, 2026
Summary
Anatomical factors like tumor location and carotid artery encasement significantly impact gross-total resection (GTR) of skull base chondrosarcomas (SBCs). An AI-powered nomogram aids in predicting GTR and optimizing surgical approach selection.
Area of Science:
- Neurosurgery
- Oncology
- Medical Imaging
- Machine Learning in Medicine
Background:
- Skull base chondrosarcomas (SBCs) present complex surgical challenges.
- Achieving gross-total resection (GTR) is crucial for optimal patient outcomes.
- Predicting the likelihood of GTR preoperatively is essential for surgical planning.
Purpose of the Study:
- To develop and validate an anatomy-driven model for predicting GTR in SBCs.
- To identify key anatomical and clinical factors influencing GTR.
- To create a reproducible tool for preoperative surgical decision-making.
Main Methods:
- Retrospective multi-institutional analysis of 185 SBCs.
- Inclusion of 13 preoperative variables (tumor location, ICA encasement, CN involvement, etc.).
- Application and comparison of five machine learning algorithms (GLM, random forest, SVM, k-NN, XGBoost) and multivariable logistic regression; nomogram development.
Main Results:
- Overall GTR rate was 56%, varying by tumor location.
- Petroclival location, ICA encasement, and lower cranial nerve involvement significantly decreased GTR odds.
- Endoscopic transpterygoid approach (ETPA) was associated with higher GTR likelihood compared to open surgery.
- The generalized linear model (GLM) achieved the best discrimination (AUC 0.838).
Conclusions:
- Anatomical factors, particularly petroclival origin, ICA encasement, and lower CN involvement, are primary barriers to SBC GTR.
- A machine learning-supported nomogram can reliably predict GTR likelihood.
- This tool aids in selecting appropriate surgical corridors and rationalizing adjuvant therapy decisions.