Related Experiment Video
Updated: Oct 5, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.9K
A multi-task two-path deep learning system for predicting the invasiveness of craniopharyngioma
Lin Zhu1, Lingling Zhang2, Wenxing Hu3
1School of Control Science and Engineering, Shandong University, Jinan, Shandong, 250061, China; CBSR&NLPR, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China.
Computer Methods and Programs in Biomedicine
|February 1, 2022
Summary
This study introduces the MT-Brain system, a novel deep learning model for predicting craniopharyngioma invasiveness and tumor boundaries from MRI scans. The system shows significant potential for improving preoperative surgical planning and patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurosurgery
Background:
- Craniopharyngioma (CP) is a benign brain tumor that can exhibit aggressive clinical behavior.
- Predicting CP invasiveness and boundaries from MRI is challenging for radiologists.
- Accurate preoperative assessment is crucial for individualized treatment and surgical planning.
Purpose of the Study:
- To develop a non-invasive deep learning method for predicting craniopharyngioma invasiveness.
- To accurately segment tumor boundaries on MRI scans.
- To provide a preoperative reference for surgical decision-making.
Main Methods:
- Developed the MT-Brain system, a two-pathway deep learning model utilizing 2D CNN and 3D sub-networks.
- Incorporated position encoding and mask-guided attention for enhanced performance.
- Trained and tested the system on 1032 craniopharyngioma patient MRI scans (302 invasive, 730 non-invasive).
Main Results:
- The MT-Brain system achieved an AUC of 83.84% for invasiveness diagnosis.
- Demonstrated high accuracy (77.94%), sensitivity (70.97%), and specificity (80.99%).
- Achieved a Dice score of 66.36% for lesion segmentation, comparable to radiologist performance.
Conclusions:
- This is the first integrated deep learning model for simultaneous preoperative prediction of CP invasiveness and lesion boundary localization.
- The MT-Brain system demonstrates excellent performance and significant potential for clinical application.
- The model's interpretability explorations support its reliability in real-world scenarios.
