Detection, Segmentation and Tracking of Cells

Status: available
SupervisorRoman Pflugfelder

There is the opportunity for a payment by

Problem Statement

Cell detection and cell tracking are important tasks in bioimaging. Bioimaging is currently radically transforming microbiology as terabytes of image data of modern microscopy needs an automatic or semi-automatic approach to efficient experimentation. With the advent of a new revolutionary gene editing technique (CRISPR/Cas), such cell experiments under the microscope are becoming easy and are now de facto standard in microbiology. Such imaging based microscopic experiments promise to observe cell development from molecular level to the whole organism simultaneously over days.

There are many challenges in microscopic bioimaging. On the one hand, cell detection and segmentation is difficult due to many factors: cells are difficult to separate as they touch each other and usually form heaps of cells. Cells have very similar appearance. On the other hand, imaging techniques such as light microscopy or fluorescent microscopy have also their problems which often yields to very low contrast, occlusion and substantial noise. Tracking of cells with modern microscopes is hard as well as typical frame-rates are in minutes rather than in seconds.

Goal

Understand  the problems of modern microscopy and understand the challenges of automatic cell detection, segmentation and tracking. Get familiar with cell data. Compose a dataset based on existing and new data, train an existing deep neural network for cell detection/segmentation and validate the results. Implement the whole method as re-usable library.

Workflow

  • Review literature
  • Create training and validation dataset
  • Implement training and test algorithms
  • Optional: extend the algorithm to cell tracking
  • Test results on new acquired cell data
  • Written report/thesis and final presentation

Requirements

  • Basic knowledge in computer vision
  • Basic experience in Matlab, C++, Python
  • Interest in Machine Learning, maths, statistics
  • Interest in GPU programming

Literature

  • C. Sommer, R. Hoefler, M. Samwer, D. W. Gerlich. A deep learning and novelty detection framework for rapid phenotyping in high-content screening. Mol Biol Cell, 28(23):3428-3436, Nov 7, 2017
  •  O. Ronneberger, P. Fischer, T. Brox. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention (MICCAI), Springer, LNCS, Vol.9351: 234–241, 2015
  • C. Feichtenhofer, A. Pinz, A. Zisserman. Detect to Track and Track to Detect. IEEE International Conference on Computer Vision (ICCV) 2017
  • K. He, G. Gkioxari, P. Dollar, R. Girshick. Mask R-CNN.  IEEE International Conference on Computer Vision (ICCV) 2017