Progressive multifocal leukoencephalopathy lesion and brain parenchymal segmentation from MRI using serial deep...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Progressive multifocal leukoencephalopathy (PML) is a severe opportunistic brain infection caused by the JC virus.
• Accurate MRI assessment of PML lesion burden and brain atrophy is crucial for disease monitoring and therapeutic response.
• Current manual quantification methods are time-consuming, subjective, and lack validation.

Purpose of the Study:

• To introduce JCnet, a fully automated method for segmenting brain parenchyma and PML lesions.
• To enable accurate and objective quantification of PML-related brain changes.
• To provide a tool for improved disease monitoring and clinical trial applications.

Main Methods:

• Developed JCnet, an end-to-end, fully automated segmentation method using 3D patch-based convolutional neural networks.
• Employed multi-view feature pyramid networks with hierarchical residual learning blocks.
• Utilized a fully convolutional architecture for output probability membership generation.

Main Results:

• JCnet demonstrates superior performance and improved longitudinal consistency compared to state-of-the-art methods.
• The method accurately segments healthy brain tissue and lesions, outperforming comparators.
• Achieved robust and accurate automated measures for brain atrophy and lesion segmentation in PML.

Conclusions:

• JCnet offers a robust and automated solution for quantifying PML brain changes.
• This method can significantly aid in clinical monitoring and the development of targeted therapies.
• Standardized quantitative MRI measures derived from JCnet hold promise for future clinical trials.