Related Experiment Video
Updated: Jul 3, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Lightweight deep learning framework for intracranial hemorrhage detection in brain CT scans
Sumaira Hussain1, Salman Jan2,3, Manal Aldhayan4
1School of Computer Science and Technology, Shandong Jianzhu University, Jinan, China.
Frontiers in Medicine
|July 2, 2026
Summary
Lightweight deep learning models show promise for detecting intracranial hemorrhage on brain CT scans. MobileNetV2 achieved 87% accuracy, offering a potential tool for faster diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Intracranial hemorrhage is a critical condition requiring rapid diagnosis via brain CT scans.
- Manual interpretation of CT scans is time-consuming and prone to inter-observer variability, potentially delaying treatment.
- Automated detection systems are needed to improve diagnostic speed and accuracy.
Purpose of the Study:
- To evaluate lightweight, mobile-optimized deep learning models for automated intracranial hemorrhage detection.
- To compare the diagnostic performance and computational efficiency of custom CNN, MobileNetV2, and EfficientNet-B0.
- To assess the feasibility of deep learning for assistive clinical decision support in neuroimaging.
Main Methods:
- Utilized a curated subset of a public brain CT hemorrhage dataset.
- Developed and trained a custom Convolutional Neural Network (CNN).
- Evaluated transfer learning models: MobileNetV2 and EfficientNet-B0 for intracranial hemorrhage detection.
Main Results:
- MobileNetV2 achieved the highest performance with 87% accuracy and 0.94 AUC.
- The custom lightweight CNN achieved 79% accuracy.
- EfficientNet-B0 showed competitive results but demanded higher computational resources.
Conclusions:
- Lightweight deep learning models can achieve reliable diagnostic performance for intracranial hemorrhage detection.
- Optimized deep learning systems show potential for supporting preliminary clinical assessment.
- Further improvements in sensitivity and extensive clinical validation are necessary for real-world deployment.