Cerebral microbleed detection using Susceptibility Weighted Imaging and deep learning.

Area of Science:

• Neuroimaging
• Medical Diagnostics
• Artificial Intelligence

Background:

• Cerebral microbleeds (CMBs) are crucial biomarkers for neurological diseases such as dementia and stroke.
• Manual CMB detection is labor-intensive and error-prone.
• Existing automated methods often generate excessive false positives.

Purpose of the Study:

• To develop and validate a novel two-stage deep learning framework for accurate cerebral microbleed detection.
• To enhance the efficiency and reliability of CMB diagnosis in clinical settings.

Main Methods:

• A two-stage approach combining 3D fast radial symmetry transform on Susceptibility Weighted Imaging (SWI) for candidate detection.
• A deep residual neural network utilizing SWI and high-pass filtered phase images for false positive reduction.
• Model training and validation using 154 and 25 datasets, respectively, with testing on 41 diverse cases.

Main Results:

• The best deep learning model achieved 95.8% sensitivity and 70.9% precision.
• The model demonstrated a low false positive rate of 1.6 per case.
• Performance was comparable to expert human raters and superior to existing automated methods.

Conclusions:

• The proposed deep learning framework significantly improves the accuracy and efficiency of cerebral microbleed detection.
• This approach holds substantial potential for advancing the diagnosis of various neurological disorders.
• Integrating deep learning with multi-modal imaging (SWI and phase images) offers a powerful tool for medical image analysis.