Math Biology Seminar: Dhananjay Bhaskar

  • Date: 10/19/2016
  • Time: 13:45
Lecturer(s):
Dhananjay Bhaskar, UBC
Location: 

University of British Columbia

Topic: 

A Machine Learning Approach to Morphology Based Cell Classification

Description: 

Individual cells regulate their morphology in response to environmental cues, selective pressures and signalling. The precise mechanism(s) through which cells control their shape is not well understood. Studies have shown that cell morphology has important implications for nutrient uptake, motility, proliferation, etc. For example, a change in bacterial cell diameter of 0.2 µm can change the energy required for chemotaxis by a factor of 10^5. Automatic classification and counting can facilitate a systematic investigation of cell morphology. Furthermore, it is a useful tool for diagnosis of diseases like leukemia that are characterized by cell shape deformation.

 

In this talk, I will describe techniques for image segmentation and feature extraction that we use to build a descriptor of cell shape. This descriptor is used to classify cells using unsupervised learning (PCA, hierarchical clustering) methods. I will briefly discuss the advantages and limitations of supervised learning (deep neural networks, convolutional neural networks) methods.

Other Information: 

Location: ESB 4127