Master presentation: Gabriel Angerd

Handledare: Daniel Midtvedt
Examinator: Giovanni Volpe
Opponent: Valdemar Bergentall

Abstract

Instance segmentation is predominately used when objects within an image is to be identified and separated from the image. This process of isolating objects within an image can be utilized in almost every field of computer vision. In this report the object is to use Mask RCNN to segment images of yeast cells under osmotic pressure. The process of introducing a different salinity to the environment of yeast cells causes osmotic pressure, effectively causing the cells to change in volume and refractive index. This process is identifiable by several methods constructed for this exact purpose. However, When the temporal change is of question many popular methods lack precision. By using Mask-RCNN to segment each individual cell within an image, another CNN-based network can quantify the physical properties of these cells. Mask-RCNN trained on simulated data efficiently and accurately segmented images of yeast with several hundred cases and we were able to identify the change in volume and refractive index caused by osmotic pressure on experimental data obtained by digital holographic microscopy. The change in volume and refractive index is proportional to the cell mass and a change of around 30% was observed using the CNN.