University of Twente Student Theses


Cell Detection in Whole Slide Images With Out-of-Focus Corruption

Sokolovas, E. (2021) Cell Detection in Whole Slide Images With Out-of-Focus Corruption.

[img] PDF
Abstract:The widespread adoption of digital whole slide scanners has allowed for the automation of laborious tasks such as cell counting in histopathological analysis using machine learning (ML) techniques. Unfortunately WSI’s tend to suffer from focus issues, degrading ML tool performance. In the literature, convolutional neural networks (CNNs) have shown positive results in cell detection and focus quality assessment tasks. So far, no paper has combined both approaches for blur resistant cell detection in WSIs. We propose a novel pipeline to address this gap. The pipeline was developed using two different CNN types: image segmentation (Unet) and object detection (Yolov4). The novel pipeline showed no appreciable performance difference in the cell counting task between a control model trained on both in-focus and out-of-focus images. The output of image segmentation models required additional processing to derive cell counts, the method used for this step was naive and provided poor results. This meant that object detection based pipeline substantially outperformed the image segmentation based pipeline. However, the control models trained on both in-focus and out-of-focus images provided overall reasonable performance indicating that some degree of robustness against blur could be achieved through the inclusion of blurry images into model training datasets.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page