Automated detection of critical findings in multi-parametric brain MRI using a system of 3D neural networks

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Brain MRI interpretation is crucial for diagnosing neurological conditions.
• Automated image classification can enhance radiologist workflow and diagnostic accuracy.
• Deep learning models show promise in medical image analysis.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for automated classification of brain MRIs.
• To assess the CNN's performance in identifying abnormal brain MRIs and critical intracranial findings: acute infarction, acute hemorrhage, and mass effect.

Main Methods:

• A CNN incorporating localization feature extraction (LFE) modules was designed.
• 13,215 clinical brain MRI studies were used, categorized into training, validation, and internal/external testing sets.
• Model performance was evaluated using Receiver Operating Characteristic (ROC) analysis, reporting AUC, sensitivity, and specificity.

Main Results:

• The CNN achieved high performance on internal test data (AUCs ranging from 0.90 to 0.95) and external test data (AUCs from 0.83 to 0.97) for detecting abnormalities and critical findings.
• Specific performance metrics varied by finding, with high accuracy for acute infarction detection.

Conclusions:

• The developed deep convolutional network accurately identifies abnormal and critical intracranial findings on brain MRIs.
• The AI tool demonstrates potential to assist neuroradiologists, improving efficiency and accuracy in brain MRI interpretation.
• The model's ability to handle varying MR contrasts addresses a practical challenge in clinical settings.