Award Date
August 2024
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science
First Committee Member
Mingon Kang
Second Committee Member
Fatma Nasoz
Third Committee Member
Kazem Taghva
Fourth Committee Member
Laxmi Gewali
Fifth Committee Member
Mira Han
Number of Pages
100
Abstract
Automatic histopathological Whole Slide Image (WSI) analysis has been highlighted along with the advancements in microscopic imaging techniques, but manual examination and diagnosis of WSIs are time-consuming and tiresome. Recently, deep convolutional neural networks have succeeded in histopathological image analysis. Especially, Convolutional Neural Networks CNN models such as Inception and DenseNet have achieved effective performance. However, automatic histopathological WSI analysis still has significant drawbacks such as considering deep learning as black-box models, predicting disease independently on a small part of images (patch images) extracted from WSIs, limitations in predicting a single slide-based score for a patient, and capturing disease-specific morphology patterns (e.g., protein rearrangements and subtype morphology) from primary screening WSI images’ protein rearrangements. To address these gaps, I have developed three interpretable and evidential deep learning models: Deep-Hipo, HipoMap, and Deep-PATHO. My proposed model explains disease specific morphology patterns which were aligned with domain experts. In Deep-Hipo, I designed multi-scale receptive fields to simultaneously capture local information (20x patch) and global information (5x). Deep-Hipo achieved significant performance over state-of-the-art models by considering patch dependencies. The Histopathology representation Map (HipoMap) is a novel and generalized slide-score prediction framework for any CNN-based patch-wise pretraining models. HipoMap generates a slide-based prediction framework by generating one representation map for a WSI. My research shows that HipoMap can be extended to many applications such as cancer classification, survival prediction, survival analysis, and sub-type classification. Deep-PATHO is a novel deep-learning architecture that synchronizes morphology between local (High magnification) and global (Low magnification). Deep-PATHO has achieved significant performance in identifying complex morphological patterns of ALK rearrangements in WSI images.
Keywords
Convolutional Neural Network; Deep Learning; Health Informatics; Interpretation; Weakly Supervised
Disciplines
Bioinformatics | Computer Sciences
Degree Grantor
University of Nevada, Las Vegas
Language
English
Repository Citation
Kosaraju, Sai Chandra, "Interpretable and Evidential Deep Learning for Medical Image Analysis" (2024). UNLV Theses, Dissertations, Professional Papers, and Capstones. 5131.
https://digitalscholarship.unlv.edu/thesesdissertations/5131
Rights
IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/