Deep learning based neuronal soma detection and counting for Alzheimer's disease analysis

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Alzheimer's Disease (AD) is characterized by neuronal damage and loss.
• Micro-Optical Sectioning Tomography (MOST) enables high-resolution whole-brain imaging for neuron analysis.
• Analyzing large-scale MOST datasets for AD progression is computationally challenging.

Purpose of the Study:

• To develop an automated method for analyzing large 3D whole-brain images acquired by MOST.
• To accurately quantify neuron numbers in Alzheimer's Disease mouse models.
• To investigate age-related changes in neuronal populations during AD progression.

Main Methods:

• Acquired 3D whole-brain MOST images from six Alzheimer's Disease mice.
• Developed a deep learning approach using Convolutional Neural Networks (CNNs) for neuronal soma detection.
• Classified image cubes into 'soma', 'fiber', or 'background' categories to count neurons.

Main Results:

• The proposed deep learning method significantly outperforms manual counting and existing software (NeuroGPS) in speed and accuracy.
• Analysis revealed a slight decrease in neuron counts in the lateral entorhinal cortex, medial entorhinal cortex, and presubiculum with age in AD mice.
• Findings align with previous experimental observations of neuronal loss in AD.

Conclusions:

• A novel, automated deep learning method effectively handles large MOST image datasets for accurate neuron identification.
• This technique facilitates the study of Alzheimer's Disease pathology by enabling comprehensive whole-brain neuron analysis.
• The method offers potential for constructing whole-brain neuron projections to visualize AD's impact on brain structure.