Modeling Neurodegeneration in silico With Deep Learning.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Neurology

Background:

• Deep neural networks (DNNs) show promise for modeling brain function.
• Current research predominantly focuses on modeling healthy brain activity.
• There is a need for computational models of neurological diseases.

Purpose of the Study:

• To propose and demonstrate a paradigm for modeling neurological diseases in silico using deep learning.
• To simulate posterior cortical atrophy (PCA), an atypical Alzheimer's disease variant, in artificial neural networks.
• To investigate the impact of simulated neural damage on visual object recognition.

Main Methods:

• Utilized deep convolutional neural networks (DCNNs) trained for visual object recognition.
• Simulated PCA by introducing random injuries to connections between artificial neurons.
• Analyzed the effects of simulated damage on network performance and learned representations.

Main Results:

• Injured DCNNs exhibited a progressive loss of object recognition capabilities.
• Simulated PCA affected learned representations hierarchically, impacting object-level before category-level recognition.
• The model provides insights into the functional consequences of neural damage in PCA.

Conclusions:

• The proposed paradigm offers a new computational approach for modeling neurological diseases.
• This method can be extended to study other cognitive domains and incorporate neural plasticity.
• Essential for developing in silico models of the brain and advancing neurological disease research.