JCnet

JCnet performs automated segmentation of Progressive Multifocal Leukoencephalopathy (PML) lesions and brain parenchyma from magnetic resonance imaging (MRI) for quantitative analysis.

Key Features:

• End-to-End Segmentation: Provides end-to-end segmentation of PML lesions and brain parenchyma from MRI data without manual delineation.
• Deep Learning Architecture: Employs 3D patch-based convolutional neural networks (CNNs) with multi-view feature pyramid networks and hierarchical residual learning blocks.
• Architectural Components: Incorporates embedded batch normalization and nonlinear activation functions within the network architecture.
• Fully Convolutional Methodology: Merges feature maps across bottom-up and top-down pathways of feature pyramids and generates output probability membership through convolutional pathways.
• Performance: Outperforms conventional state-of-the-art methods used for healthy brain and multiple sclerosis lesion segmentation and demonstrates robustness in longitudinal studies.
• Quantitative MRI Measures: Produces standardized quantitative measures of lesion burden and brain parenchymal volume for clinical assessment and trials.
• Serial MRI Processing: Capable of processing serial MRI data for longitudinal analysis.

Scientific Applications:

• Clinical Monitoring: Quantitative assessment of PML lesion burden and brain parenchymal volume to monitor disease progression and response to therapy.
• Clinical Trials: Generation of standardized quantitative MRI endpoints for trials of targeted therapies in PML.
• Neuroimaging Research: Objective segmentation for comparative studies of lesion patterns in PML, healthy brain, and multiple sclerosis.

Methodology:

Implemented in Python 3 using Keras and TensorFlow; employs 3D patch-based CNNs with multi-view feature pyramid networks, hierarchical residual learning blocks, embedded batch normalization, nonlinear activations, fully convolutional merging of bottom-up and top-down pathways, and produces probabilistic output membership for segmentation of serial MRI data.