Related Experiment Video
Updated: Aug 6, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Explainable brain tumor detection in skull base CT using continuous neural representations: comparative evaluation of
Hilal Akarkamçı1, Özge Demir2, Koray Acici3
1Ministry of National Education, Directorate General for Higher and Foreign Education, Ankara, Turkey.
Neuroscience
|July 30, 2026
Summary
Neural Fields, a continuous neural representation learning method, significantly improves brain tumor detection accuracy on CT scans. This approach offers superior performance and interpretability over transformer and operator models for challenging skull base analyses.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Brain tumor detection in CT scans is difficult due to low contrast, complex skull base anatomy, and imaging artifacts.
- Continuous neural representation learning offers a potential solution for improving diagnostic performance and interpretability.
Purpose of the Study:
- To evaluate the effectiveness of continuous neural representation learning for CT-based brain tumor detection.
- To compare the performance of various neural models, including neural fields, neural operators, and transformers, on a skull base CT dataset.
Main Methods:
- A dataset of 200 patients (40,000 slices) from Gazi University Faculty of Medicine was used.
- An automated multi-stage slice selection framework extracted three key skull base slices per patient.
- Ten models, including Neural Fields, Implicit Neural Representations, DeepONet, and various transformers (ViT), were evaluated.
Main Results:
- Neural Fields achieved the highest accuracy (98.90%), sensitivity (98.90%), and specificity (98.90%).
- Neural Fields demonstrated superior performance compared to transformer and operator-based models, with statistically significant differences.
- LIME analysis confirmed Neural Fields focused on relevant lesion areas and reduced artifact influence.
Conclusions:
- Continuous neural representations, specifically Neural Fields, offer enhanced accuracy and robustness for brain tumor classification in CT scans.
- Neural field-based learning provides high diagnostic performance and improved interpretability, supporting its clinical use for decision support in skull base CT analysis.
