Fluorescence microscopy image classification of 2D HeLa cells based on the CapsNet neural network

Area of Science:

• Cell Biology
• Computational Biology
• Microscopy

Background:

• Fluorescence microscopy generates large image datasets for protein localization.
• Digital image processing aids in classifying protein images and identifying subcellular locations.
• Understanding protein function relies on accurate subcellular localization data.

Purpose of the Study:

• To apply the CapsNet network model for classifying protein images in different subcellular compartments.
• To evaluate the accuracy of CapsNet in analyzing 2D fluorescence microscopy datasets of HeLa cells.

Main Methods:

• Utilized a 2D image dataset of HeLa cells from fluorescence microscopy.
• Implemented the CapsNet network model, focusing on capsule training to capture feature possibilities and variants.
• Employed dynamic routing for capsule activation based on consistent predictions.

Main Results:

• The CapsNet model demonstrated high accuracy in classifying ten types of protein images across different subcellular compartments.
• Capsules were trained to generalize feature detection rather than memorizing specific instances.
• Dynamic routing mechanism enabled robust predictions through consensus among capsules.

Conclusions:

• CapsNet is an effective deep learning model for accurate protein subcellular localization using fluorescence microscopy images.
• The study highlights the potential of CapsNet in advancing high-throughput biological image analysis.
• This approach facilitates more efficient investigation of protein function through precise localization.