An artificial immune-activated neural network applied to brain 3D MRI segmentation

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Biomedical Imaging

Background:

• Accurate segmentation of 3D brain MRI data is crucial for medical diagnosis and research.
• Existing segmentation methods face challenges, especially with noisy image data.
• Biologically inspired computing offers novel approaches to complex pattern recognition tasks.

Purpose of the Study:

• To introduce a new neural network model, the Artificial Immune-Activated Neural Network (AIANN).
• To apply the AIANN to the segmentation of 3D brain MRI data.
• To evaluate the performance of the AIANN against established methods.

Main Methods:

• Developed an AIANN model with neuron activation functions inspired by biological immune system interactions (receptor-epitope binding).
• Incorporated an energy measure to control recognition accuracy within the AIANN.
• Applied the AIANN to segment real and simulated 3D brain MRI datasets.

Main Results:

• The AIANN model demonstrated superior segmentation accuracy compared to published methods on both real and simulated MRI data.
• The model showed particular effectiveness in segmenting brain structures even at low noise levels.
• Evaluation utilized datasets from Massachusetts General Hospital and the McGill University BrainWeb simulator.

Conclusions:

• The AIANN presents a promising new approach for medical image segmentation, leveraging principles from the immune system.
• The model's robust performance, especially in noisy environments, highlights its potential clinical and research applications.
• Further research can explore AIANN's adaptability to other neuroimaging modalities and segmentation tasks.