Generative Model of Brain Microbleeds for MRI Detection of Vascular Marker of Neurodegenerative Diseases

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Cerebral microbleeds (CMB) are associated with aging, vascular pathologies, and neurodegeneration.
• Accurate detection of CMB using deep learning on MRI is hindered by limited ground truth data and confounding imaging features like vessels and infarcts.

Purpose of the Study:

• To develop a novel generative adversarial network (GAN) for generating realistic 3D cerebral microbleed (CMB) lesions.
• To create large, diverse training datasets for deep learning-based CMB detection models.
• To investigate CMB characteristics and improve diagnostic accuracy.

Main Methods:

• A novel 3D lesion generative adversarial network (GAN) was developed and trained to generate synthetic CMB conditioned by volume and location.
• The generated synthetic CMB data was used to train a convolutional neural network (CNN) classifier.
• The model's performance was evaluated on real CMB detection and its applicability to unseen datasets with varying MRI parameters and diseases was assessed.

Main Results:

• Achieved state-of-the-art performance in detecting real cerebral microbleeds (CMB) using a CNN classifier trained on synthetic CMB data.
• Demonstrated the 3D lesion GAN's ability to generate diverse synthetic lesions on unseen datasets without requiring manual ground truth annotation.
• Validated the model's robustness across different MRI parameters and disease contexts.

Conclusions:

• The proposed 3D lesion GAN effectively generates synthetic CMB, significantly improving deep learning-based detection.
• This method addresses the challenge of limited annotated data, enabling the creation of large training datasets.
• The GAN model shows broad applicability for generating diverse synthetic lesions in neuroimaging research and diagnostics.