{"id":{"repo_id":"gatech","oai_identifier":"oai:repository.gatech.edu:1853/78670"},"canonical_url":"https://search.dev.ndltd.org/etd/gatech/oai:repository.gatech.edu:1853/78670","repository":{"repo_id":"gatech","name":"Georgia Tech","base_url":"https://repository.gatech.edu/server/oai/request"},"display":{"title":"Knowledge-Informed Weakly-Supervised Deep Learning Models for Cancer Applications","abstract":"In recent decades, deep learning (DL) has emerged as a powerful tool for analyzing complex patterns in large-scale healthcare data, significantly advancing diagnosis, prognosis, and treatment planning. However, the collection of medical data faces inherent limitations, including invasiveness, high cost, expert labeling requirements, disease rarity, and patient recruitment challenges. This thesis addresses these constraints by developing novel knowledge-informed, image-based DL methodologies that enhance sample efficiency, predictive accuracy, and generalizability for real-world cancer applications. The proposed models systematically integrate biological, anatomical, and clinical domain knowledge into DL pipelines to overcome data scarcity and heterogeneity in cancer imaging. They combine self-supervised pretraining, knowledge-informed loss functions, hierarchical and contextualized architectures, and label smoothing techniques that distill clinical and biological priors. These approaches enable dense spatial prediction of gene modules, genetic alterations, and segmentation of heterogeneous tumors with vague boundaries, even under sparse supervision. Across applications in glioblastoma and liver cancer, the methods demonstrate substantial improvements in generalizability and precision, showing strong potential to support personalized diagnosis, prognosis, treatment planning, and monitoring in precision oncology.","abstract_html":"In recent decades, deep learning (DL) has emerged as a powerful tool for analyzing complex patterns in large-scale healthcare data, significantly advancing diagnosis, prognosis, and treatment planning. However, the collection of medical data faces inherent limitations, including invasiveness, high cost, expert labeling requirements, disease rarity, and patient recruitment challenges. This thesis addresses these constraints by developing novel knowledge-informed, image-based DL methodologies that enhance sample efficiency, predictive accuracy, and generalizability for real-world cancer applications. The proposed models systematically integrate biological, anatomical, and clinical domain knowledge into DL pipelines to overcome data scarcity and heterogeneity in cancer imaging. They combine self-supervised pretraining, knowledge-informed loss functions, hierarchical and contextualized architectures, and label smoothing techniques that distill clinical and biological priors. These approaches enable dense spatial prediction of gene modules, genetic alterations, and segmentation of heterogeneous tumors with vague boundaries, even under sparse supervision. Across applications in glioblastoma and liver cancer, the methods demonstrate substantial improvements in generalizability and precision, showing strong potential to support personalized diagnosis, prognosis, treatment planning, and monitoring in precision oncology.","abstract_has_math":false,"creators":["Wang, Hairong"],"institution":"Georgia Institute of Technology","degree_name":null,"degree_level":"Doctoral","degree_discipline":null,"degree_department":"Industrial and Systems Engineering","school":null,"contributors":[],"advisors":["Li, Jing"],"committee_chairs":[],"committee_members":["Shi, Jianjun","Xian, Xiaochen","Swanson, Kristin","Huang, Shuai"],"year":2025,"date_issued":"2025-06-23","date_published":"2025-06-23","updated_at":"2026-07-27T19:51:43Z","subjects":["Deep learning","Knowledge-informed models","Weakly supervised learning"],"languages":["en_US"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1853/78670","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Li, Jing"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Shi, Jianjun","Xian, Xiaochen","Swanson, Kristin","Huang, Shuai"]},{"key":"dc:contributor.department","label":"Department","values":["Industrial and Systems Engineering"]},{"key":"dc:creator","label":"Author","values":["Wang, Hairong"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-08-15T12:41:10Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-08-15T12:41:10Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-06-23"]},{"key":"dc:publisher","label":"Institution","values":["Georgia Institute of Technology"]},{"key":"dc:type","label":"Dc Type","values":["Text"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Doctoral"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Deep learning","Knowledge-informed models","Weakly supervised learning"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en_US"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1853/78670"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["In recent decades, deep learning (DL) has emerged as a powerful tool for analyzing complex patterns in large-scale healthcare data, significantly advancing diagnosis, prognosis, and treatment planning. However, the collection of medical data faces inherent limitations, including invasiveness, high cost, expert labeling requirements, disease rarity, and patient recruitment challenges. This thesis addresses these constraints by developing novel knowledge-informed, image-based DL methodologies that enhance sample efficiency, predictive accuracy, and generalizability for real-world cancer applications. The proposed models systematically integrate biological, anatomical, and clinical domain knowledge into DL pipelines to overcome data scarcity and heterogeneity in cancer imaging. They combine self-supervised pretraining, knowledge-informed loss functions, hierarchical and contextualized architectures, and label smoothing techniques that distill clinical and biological priors. These approaches enable dense spatial prediction of gene modules, genetic alterations, and segmentation of heterogeneous tumors with vague boundaries, even under sparse supervision. Across applications in glioblastoma and liver cancer, the methods demonstrate substantial improvements in generalizability and precision, showing strong potential to support personalized diagnosis, prognosis, treatment planning, and monitoring in precision oncology."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Ph.D."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Knowledge-Informed Weakly-Supervised Deep Learning Models for Cancer Applications"]}]}],"canonical_facts":{"dc:contributor.advisor":["Li, Jing"],"dc:contributor.committeemember":["Shi, Jianjun","Xian, Xiaochen","Swanson, Kristin","Huang, Shuai"],"dc:contributor.department":["Industrial and Systems Engineering"],"dc:creator":["Wang, Hairong"],"dc:date.accessioned":["2025-08-15T12:41:10Z"],"dc:date.available":["2025-08-15T12:41:10Z"],"dc:date.issued":["2025-06-23"],"dc:description.abstract":["In recent decades, deep learning (DL) has emerged as a powerful tool for analyzing complex patterns in large-scale healthcare data, significantly advancing diagnosis, prognosis, and treatment planning. However, the collection of medical data faces inherent limitations, including invasiveness, high cost, expert labeling requirements, disease rarity, and patient recruitment challenges. This thesis addresses these constraints by developing novel knowledge-informed, image-based DL methodologies that enhance sample efficiency, predictive accuracy, and generalizability for real-world cancer applications. The proposed models systematically integrate biological, anatomical, and clinical domain knowledge into DL pipelines to overcome data scarcity and heterogeneity in cancer imaging. They combine self-supervised pretraining, knowledge-informed loss functions, hierarchical and contextualized architectures, and label smoothing techniques that distill clinical and biological priors. These approaches enable dense spatial prediction of gene modules, genetic alterations, and segmentation of heterogeneous tumors with vague boundaries, even under sparse supervision. Across applications in glioblastoma and liver cancer, the methods demonstrate substantial improvements in generalizability and precision, showing strong potential to support personalized diagnosis, prognosis, treatment planning, and monitoring in precision oncology."],"dc:description.degree":["Ph.D."],"dc:format.mimetype":["application/pdf"],"dc:identifier.uri":["https://hdl.handle.net/1853/78670"],"dc:language.iso":["en_US"],"dc:publisher":["Georgia Institute of Technology"],"dc:subject":["Deep learning","Knowledge-informed models","Weakly supervised learning"],"dc:title":["Knowledge-Informed Weakly-Supervised Deep Learning Models for Cancer Applications"],"dc:type":["Text"],"thesis:degree_level":["Doctoral"]},"updated_at":"2026-07-27T19:51:43Z"}