The aim of this thesis is to develop an efficient method for tracking different cellular structures in microscopy image sequences using segmentation and tracking algorithms. The problem of cell tracking in microscopy images involves over-segmentation and under-segmentation, which makes the linking of sequential detections alone insufficient for successful tracking. For this purpose, we used the Cellpose model for cell segmentation, which is complemented by a tracker propagating the detected cells through a sequence of images, developed based on vector gradients and additional post-processing methods to improve accuracy. Multi-frame information is also used to re-find cells in the event of temporary mutual occlusion. The method we developed proved to be effective in dealing with cases where cells are too close; using gradient information, we successfully separated cells in close contact. Multi-frame tracking also preserves cell trajectories even in cases of occlusion, allowing more accurate monitoring of cells over longer periods of time. The results showed an improvement in cell recognition under challenging conditions, such as edge-on cases or cell clusters. In the field of cell tracking, simple methods still outperform deep methods, and our research contributes to the development of more reliable methods for cell tracking and opens the way for further improvements in microscopic analysis.
|
