A Hierarchical Convolutional Neural Network for vesicle fusion event classification.

Area of Science:

• Biophysics
• Cell Biology
• Machine Learning

Background:

• Quantitative analysis of vesicle exocytosis is crucial for understanding cellular processes.
• Classifying different vesicle fusion modes from fluorescence microscopy data presents significant challenges.

Purpose of the Study:

• To develop an automated method for identifying and classifying vesicle fusion events.
• To improve the accuracy and efficiency of vesicle exocytosis analysis using machine learning.

Main Methods:

• A novel Hierarchical Convolutional Neural Network (HCNN) was proposed for analyzing time-lapse Total Internal Reflection Fluorescence Microscopy (TIRFM) image sequences.
• A detection and tracking method was employed to extract relevant image patches, followed by Gaussian Mixture Model (GMM) fitting.
• The HCNN integrated time-series intensity features from GMM and visual appearance features for classification.

Main Results:

• The HCNN method successfully classified candidate image sequences into three categories: full fusion, partial fusion, and non-fusion events.
• Performance validation on nine challenging, cell biologist-annotated datasets demonstrated superior results compared to three existing methods.

Conclusions:

• The proposed HCNN offers a robust and accurate automated approach for quantitative analysis of vesicle fusion.
• This method has the potential to significantly aid biomedical research in studying exocytosis and related cellular dynamics.