Intelligent Medical Image Analysis for Quality Assurance, Teaching and Evaluation
Date
2020-06-23
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Manually spotting and annotating the affected area(s) on histopathological images with high accuracy is regarded as the gold standard in cancer diagnosis and grading. However, this is a time-consuming and tedious task that requires considerable effort, expertise and experience of a pathologist. These are gained over time by analyzing more cases. Whereas this visual interpretation has strict guidelines. This brings a certain subjectivity to the histological analysis, and therefore, leads to inter/intra-observer variability and some reproducibility issues. Besides, these issues may have a direct effect on patient prognosis and treatment plan. These problems can be alleviated by developing automated image analysis tools for digitized histopathology. Thanks to the rapid development in the image capturing and analysis technology which could be employed to not only give more insight to pathologists, but also guide them in detecting and grading diseases. These quantitative computational tools aim to improve the quality of pathology researchers in terms of speed and accuracy. Thus, it is very important to develop an automatic assessment tool for quantitative and qualitative analysis to help remove this drawback. The main contribution of this thesis is an intelligent system for quality assurance, teaching and evaluation applications in anatomical pathology. We present a spatial clustering algorithm, named CutESC (Cut-Edge for Spatial Clustering) with a graph-based approach. CutESC performs clustering automatically for complicated shapes and different density without requiring any prior information and parameters. We have developed an automatic cell nuclei detection method where the proposed solution uses the traditional CNN learning scheme solely to detect nuclei, and then applies single-pass voting with spatial clustering explicitly to detect them. We also propose an automated method to identify and locate the mitotic cells, and tubules in histopathology images using deep neural network frameworks. We present a dataset of breast cancer histopathology images named BreCaHAD which is publicly available to the biomedical imaging community. Moreover, we propose an efficient method for salient region detection. Finally, we introduce a new tool called CACTUS (Cancer Image Annotating, Calibrating, Testing, Understanding and Sharing) which is proposed to help and guide pathologists in their effort to improve disease diagnosis and thereby reduce their workload and bias among them. CACTUS can be useful for both disseminating anatomical pathology images for teaching, as well as for evaluating agreement amongst pathologists or against a gold standard for evaluation or quality assurance.
Description
Keywords
medical image analysis, histopathology, breast cancer, proximity graph, clustering, data mining, machine learning, deep learning
Citation
Aksac, A. (2020). Intelligent Medical Image Analysis for Quality Assurance, Teaching and Evaluation (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.