Biomarkers.

Area of Science:

• Neuroradiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Microbleed detection is crucial for monitoring amyloid-related imaging abnormalities (ARIA) severity in anti-amyloid therapy (AAT).
• Susceptibility-weighted imaging (SWI) is more sensitive to microbleeds than T2*/GRE sequences.
• Manual microbleed assessment is time-consuming and subject to reader variability, necessitating automated solutions.

Purpose of the Study:

• To develop and validate a deep learning-based automated system for microbleed detection using SWI MRI.
• To improve the efficiency and reliability of ARIA-H assessment in clinical practice.

Main Methods:

• An Attention U-Net architecture with deep supervision was trained on 565 SWI MRI scans.
• The dataset included 429 positive and 136 negative cases, with microbleeds labeled by an experienced neuroradiologist.
• Model performance was validated using Dice coefficient and lesion-level Matthews correlation coefficient (MCC).

Main Results:

• The automated system achieved an AUC of 0.872, with a sensitivity of 0.677 and specificity of 0.893 on 114 test scans.
• The model detected 146 out of 158 microbleeds, with minimal impact on ARIA-H severity classification.
• Patient-level analysis showed 1.28 microbleeds per scan and 1.06 false positives per scan.

Conclusions:

• A robust automated microbleed detection approach using SWI MRI was developed, aiding ARIA-H diagnosis and severity categorization.
• The system enhances the efficiency and reliability of microbleed detection for anti-amyloid therapy monitoring.
• Future research will focus on multi-center validation and incorporating additional ARIA-related factors for comprehensive assessment.