Deep Phenotypic Cell Classification using Capsule Neural Network.

Area of Science:

• Cell biology
• Microscopy
• Machine learning

Background:

• Ultra-high-throughput microscopy enables image-based cell classification and morphological profiling.
• Current single-cell analysis methods are slow and require costly genetic/epigenetic analysis.
• Cellular heterogeneity plays a significant role in biological diversity.

Purpose of the Study:

• To propose an innovative deep learning (DL) system for accurate cell classification.
• To leverage capsule networks (CapsNet) for enhanced feature extraction in cell images.
• To provide a cost-effective and efficient alternative to existing cell analysis techniques.

Main Methods:

• Development of a deep learning system utilizing capsule networks (CapsNet).
• CapsNet architecture employs 'Capsules' for high-level feature abstraction relevant to cell categories.
• Classification of cells based on phenotypic label-free bright-field images.

Main Results:

• The proposed deep CapsNet system achieved over 98.06% accuracy in classifying different cell types.
• Deep CapsNet models demonstrated superior performance compared to conventional Convolutional Neural Network (CNN) models.
• The system effectively classifies cells using only phenotypic, label-free bright-field images.

Conclusions:

• The developed deep CapsNet system offers a highly accurate and efficient method for cell classification.
• Capsule networks provide an effective approach for high-level feature abstraction in image-based cell analysis.
• This label-free, image-based approach presents a significant advancement over traditional, costly single-cell analysis techniques.