Deep Learning to Decipher the Progression and Morphology of Axonal Degeneration

Area of Science:

• Neuroscience
• Computational Biology
• Pathology

Background:

• Axonal degeneration (AxD) is a key feature in neurodegenerative diseases, necessitating better understanding of its mechanisms.
• Identifying morphological patterns of AxD is crucial for developing targeted therapies.

Purpose of the Study:

• To evaluate the progression of AxD in cortical neurons using a novel microfluidic device and deep learning.
• To develop a high-throughput analysis tool for enhanced detection of AxD features in microscopic images.

Main Methods:

• Development of a deep learning tool (CNN) for sensitive and specific segmentation of axons, axonal swellings, and fragments.
• Utilizing a microfluidic device for in vitro modeling of AxD in hemorrhagic stroke induced by hemin.
• Application of a recurrent neural network (RNN) to identify distinct morphological patterns of AxD.

Main Results:

• The CNN tool demonstrated superior performance compared to human evaluators in segmenting AxD features.
• A time-dependent decrease in axon area and increase in axonal swelling and fragment areas were observed.
• Axonal swellings were identified as potential early predictors of axon fragmentation.
• Four distinct morphological patterns of AxD (granular, retraction, swelling, transport degeneration) were identified, indicating heterogeneity.

Conclusions:

• The developed EntireAxon platform enables systematic, time-lapse analysis of axons and AxD.
• Findings reveal morphological heterogeneity in AxD during hemorrhagic stroke, offering new insights into disease mechanisms.
• The study highlights the potential of deep learning in advancing neurodegenerative disease research.