Related Experiment Video
Updated: Jun 26, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.4K
Towards Robust Brain Midline Shift Detection: A YOLO-Based 3D Slicer Extension with a Novel Dataset
Meltem Kurt Pehlivanoğlu1, Nur Banu Albayrak2, Deniz Karhan3
1Department of Computer Engineering, Kocaeli University, Kocaeli, 41001, Turkey.
Neuroinformatics
|October 14, 2025
Summary
This study introduces a new dataset and a 3D Slicer extension for detecting brain midline shift. YOLOv5m was the optimal deep learning model, enhancing AI-assisted medical imaging for neurological condition assessment.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neuroscience
Background:
- Accurate detection of brain midline shift is crucial for diagnosing and monitoring neurological conditions like traumatic brain injuries, strokes, and tumors.
- Existing tools and datasets for brain midline shift detection are limited, hindering accurate assessment.
- This research addresses the need for improved automated detection methods in neuroimaging.
Purpose of the Study:
- To develop and evaluate deep learning models for automatic brain midline shift detection using MRI scans.
- To introduce a novel, annotated dataset (brain-midline-detection) for identifying key brain landmarks (Anterior Falx, Posterior Falx, Septum Pellucidum).
- To create a 3D Slicer extension for streamlined, AI-assisted midline shift analysis.
Main Methods:
- A comprehensive performance evaluation of deep learning models including YOLOv5 (n, s, m, l), YOLOv8, and YOLOv9 (GELAN-C).
- Development of the brain-midline-detection dataset for landmark identification in MRI scans.
- Integration of the best-performing model into a 3D Slicer extension, including preprocessing, filtering, skull stripping, registration, and shift computation.
Main Results:
- YOLOv5l achieved the highest precision (0.9601) and recall (0.9489).
- YOLOv5m demonstrated the best mAP@0.5:0.95 score (0.6087) and was selected as the optimal model due to balanced performance.
- YOLOv8s showed a higher mAP@0.5:0.95 score (0.6382) but was less practical due to high loss values. YOLOv9-GELAN-C performed poorly.
Conclusions:
- The developed 3D Slicer extension with the YOLOv5m model provides an effective tool for automated brain midline shift detection.
- The introduction of the brain-midline-detection dataset and open-source tools facilitates advancements in AI-assisted neuroimaging.
- This work contributes to more accurate, efficient, and accessible medical imaging analysis for brain midline assessment.
