Author ORCID Identifier
Semester
Fall
Date of Graduation
2024
Document Type
Dissertation
Degree Type
PhD
College
Statler College of Engineering and Mineral Resources
Department
Lane Department of Computer Science and Electrical Engineering
Committee Chair
Donald A. Adjeroh
Committee Member
Gianfranco Doretto
Committee Member
Jeremy Dawson
Committee Member
Xin Li
Committee Member
Ivan Martinez
Abstract
In recent decades, deep learning approaches have shown significant improvement in various image understanding tasks. However, analysis of high-resolution images remains a major challenge. In this work, we address the challenge of very high-resolution histopathological image (VHRHI) classification using a new information-theoretic discriminative patch selection approach. We show results on a high-resolution image dataset, namely, gigapixel whole slide tissue images for cancer tumors. Then we address how to efficiently classify challenging histopathology images, such as gigapixel whole-slide images for cancer diagnostics with image-level annotation. These ``weak labels'' are applied throughout the image but describe tumor regions of variable sizes and may refer to an extreme minority of pixels. This creates an important problem during patch-level classification, where the majority of patches in an image labeled 'cancerous' are actually tumor-free. In next work, we further improve on the problem of having very tiny tumor regions inside the image. We add a group-based assignment model to the framework pipeline. Improve the information-theoretic cluster-based sampling approach by incorporating cluster distribution information in the algorithm. Furthermore, by analyzing the spatial neighborhood of the predicted positive patches, we determine the image-level labels from the patch-level predictions. Our result based on different WSIs databases shows that the proposed approaches can perform quite competitively using a low proportion of the original dataset. Additionally, we propose a framework for efficient processing of very high-resolution aerial images (VHRAI), involving potentially megapixel dimension imagery. The framework uses the visual transformer (ViT) as its backbone and makes a decision about the scene by considering information from small tokens. Lastly, we present a novel patch-based model trustworthiness evaluation framework for very high-resolution histopathology images where the trustworthiness framework considers the rationales used by the machine to make its decisions.
Recommended Citation
Nouyed, Mohammad I., "Efficient Classification of Very High Resolution Images" (2024). Graduate Theses, Dissertations, and Problem Reports. 12699.
https://researchrepository.wvu.edu/etd/12699
Included in
Artificial Intelligence and Robotics Commons, Bioinformatics Commons, Biomedical Informatics Commons, Databases and Information Systems Commons, Data Science Commons, Diagnosis Commons, Medical Pathology Commons, Pathology Commons
