Cerebral Microbleed Automatic Detection System Based on the "Deep Learning".

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Cerebral microbleeds (CMBs) are indicative of cerebral small vessel disease (CSVD).
• Accurate detection and segmentation of CMBs are crucial for clinical diagnosis and management.
• Existing methods for CMB analysis can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To validate a deep learning system (DLS) for reliable and efficient automatic segmentation of CMBs.
• To assess the clinical diagnostic performance of the DLS in practice.
• To establish the DLS as a trusted tool for CMB detection.

Main Methods:

• A retrospective study utilized Magnetic Resonance Imaging-Susceptibility Weighted Imaging (MRI-SWI) data from 1,615 patients with CSVD and CMBs.
• A three-dimensional convolutional neural network (CNN) was trained and validated on a large dataset.
• The DLS performance was evaluated on an independent cohort of 72 patients and compared against neuroradiologist assessments.

Main Results:

• The DLS achieved a Dice coefficient of 0.72 for CMB detection and segmentation.
• Neuroradiologists confirmed that the DLS directly detected over 90% of lesions in an independent clinical dataset.
• The DLS demonstrated high agreement with expert evaluations, with an average interrater agreement kappa value of 0.79.

Conclusions:

• Automatic detection and segmentation of CMBs using the developed DLS are feasible and reliable.
• The well-trained DLS shows potential as a trusted tool for clinical application in identifying CMB lesions.
• This automated approach can enhance the efficiency and consistency of CMB diagnosis in patients with CSVD.