Columbia Technology Ventures

Classification of macrophages using deep convolutional neural networks

This technology is a classification system that can identify different types of macrophages from microscopy images using deep convolutional neural networks (CNNs).

Unmet Need: Simple method for characterization of macrophages

Macrophages perform different homeostatic activities and change their physiology depending on their environment, and as a result, have been classified into the M1 and M2 phenotypes. However, differentiating between the two types is not sufficiently obvious to researchers as macrophages can exhibit complex or mixed phenotypes. In addition, current methods for the characterization of macrophages, which usually include cell staining or fluorescence imaging, are time consuming, labor intensive, and expensive.

The Technology: High-throughput macrophage classification system using deep CNNs

This technology uses transfer learning to create an efficient macrophage phenotype classification method. Applying the CNN AlexNet, which has proven robustness in classifying diverse inputs, the program was trained to differentiate between macrophages of known phenotypes in phase contrast microscopy images. As a result, the resource cost of macrophage classification is reduced and accuracy is enhanced compared to conventional methods. While this technology can be used to further study macrophage response to stimuli, it can also be trained to differentiate between various cell types and extended to create a universal cell classification software.

This technology has been validated with macrophages that were differentiated and polarized from human THP-1 monocyte-like cell cultures.

Applications:

  • Macrophage classification
  • Live cell identification and tracking in combination with phase contrast microscopy
  • Building a universal cell classification software
  • Detection of cancer cells or genetic disorders
  • Research tool for cell identification and tracking

Advantages:

  • Does not require time-consuming and labor-intensive staining protocols
  • Does not require chemical analysis
  • Minimizes wasting of cells for characterization
  • Compatible with existing phase contrast microcopy methods
  • Cost-effective
  • Can be applied to various species and cell types
  • Has a higher learning capacity compared to standard neural networks

Lead Inventor:

Helen Lu, Ph.D.

Tech Ventures Reference: