Title A new multiresolution deep convolutional neural network workflow for glomeruli segmentation using an iterative annotation strategy /
Translation of Title Nauja daugialypės raiškos giliuoju sąsukos neuroniniu tinklu pagrįsta glomerulų segmentavimo procedūra taikant iteracinę anotavimo strategiją.
Authors Taiwo, Oluwaseun Ezekiel
Full Text Download
Pages 78
Keywords [eng] Glomeruli segmentation, Whole slide image, Deep convolutional neural networks, Iterative Annotation, Multiresolution attention residual UNET, ensemble models
Abstract [eng] One of the major approaches engaged by kidney pathologist to knowing the morphological states of the kidney glomeruli in any glomeruli disease diagnosis is image analysis using image segmentation techniques. There were attempts using various deep convolutional neural network workflow to automate glomeruli semantic segmentation in kidney images but not still achieve the most accurate results. In this current research, a new multiresolution deep convolutional neural network workflow is developed. It applies an iterative annotation strategy which involves the customization processing units of deep convolutional neural network architecture. It optimizes kidney glomeruli semantic segmentation accuracy in whole slide images. The result from this study showed that a multiresolution attention residual UNET model achieves the best segmentation accuracy of 89.89% with ensemble models.
Dissertation Institution Vilniaus universitetas.
Type Master thesis
Language English
Publication date 2022