{"id":{"repo_id":"njit","oai_identifier":"oai:digitalcommons.njit.edu:theses-4579"},"canonical_url":"https://search.ndltd.org/etd/njit/oai:digitalcommons.njit.edu:theses-4579","repository":{"repo_id":"njit","name":"NJIT","base_url":"https://digitalcommons.njit.edu/do/oai/"},"display":{"title":"Robustness of ai-driven histopathology under real-world adversarial examples","abstract":"<p>This robustness of histopathology classification models under adversarial and real-world perturbations resembling clinical artifacts is being investigated.</p><p>Using whole-slide images from the CAMELYON17 cohort, four representative architectures—ResNet-18, ResNet-50, HIPT-2MLP, and ViT-B/16 —are benchmarked across controlled pixel-level distortions and artifact-like transformations. Adversarial methods include iterative Fast Gradient Sign, Projected Gradient Descent, Salt-and-Pepper noise, and the Adversarial Watermark—Stain Shift (AWSS). Three defense strategies—Randomized Smoothing, Adversarial Training, and an Artifact Detector—are evaluated for their ability to preserve diagnostic accuracy and model reliability. Structured perturbations consistently degrade performance, with transformer-based models showing the greatest sensitivity. The benchmark developed here offers a reproducible framework for assessing robustness at both tile and slide levels, supporting safer deployment of histopathology AI in clinical workflows.</p>","abstract_html":"&lt;p&gt;This robustness of histopathology classification models under adversarial and real-world perturbations resembling clinical artifacts is being investigated.&lt;/p&gt;&lt;p&gt;Using whole-slide images from the CAMELYON17 cohort, four representative architectures—ResNet-18, ResNet-50, HIPT-2MLP, and ViT-B/16 —are benchmarked across controlled pixel-level distortions and artifact-like transformations. Adversarial methods include iterative Fast Gradient Sign, Projected Gradient Descent, Salt-and-Pepper noise, and the Adversarial Watermark—Stain Shift (AWSS). Three defense strategies—Randomized Smoothing, Adversarial Training, and an Artifact Detector—are evaluated for their ability to preserve diagnostic accuracy and model reliability. Structured perturbations consistently degrade performance, with transformer-based models showing the greatest sensitivity. The benchmark developed here offers a reproducible framework for assessing robustness at both tile and slide levels, supporting safer deployment of histopathology AI in clinical workflows.&lt;/p&gt;","abstract_has_math":false,"creators":["Yajnik, Ruchik N."],"institution":null,"degree_name":"Master of Science in Computer Science - (M.S.)","degree_level":null,"degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":["Cong Shi","Lijing Wang","Xuan Liu"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T08:44:11Z","subjects":["Machine learning","Adversarial machine learning","Histopathology","Digital pathology","Medical imaging","Bioimaging and Biomedical Optics","Computer Sciences"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.njit.edu/theses/3555","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.njit.edu/theses/3555","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://digitalcommons.njit.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.njit.edu%3Atheses-4579","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Cong Shi","Lijing Wang","Xuan Liu"]},{"key":"dc:creator","label":"Author","values":["Yajnik, Ruchik N."]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.created","label":"Dc Date Created","values":["2026-05-31T07:00:00Z"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science in Computer Science - (M.S.)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine learning","Adversarial machine learning","Histopathology","Digital pathology","Medical imaging","Bioimaging and Biomedical Optics","Computer Sciences"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.njit.edu/theses/3555"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["<p>This robustness of histopathology classification models under adversarial and real-world perturbations resembling clinical artifacts is being investigated.</p><p>Using whole-slide images from the CAMELYON17 cohort, four representative architectures—ResNet-18, ResNet-50, HIPT-2MLP, and ViT-B/16 —are benchmarked across controlled pixel-level distortions and artifact-like transformations. Adversarial methods include iterative Fast Gradient Sign, Projected Gradient Descent, Salt-and-Pepper noise, and the Adversarial Watermark—Stain Shift (AWSS). Three defense strategies—Randomized Smoothing, Adversarial Training, and an Artifact Detector—are evaluated for their ability to preserve diagnostic accuracy and model reliability. Structured perturbations consistently degrade performance, with transformer-based models showing the greatest sensitivity. The benchmark developed here offers a reproducible framework for assessing robustness at both tile and slide levels, supporting safer deployment of histopathology AI in clinical workflows.</p>"]},{"key":"dc:title","label":"Title","values":["Robustness of ai-driven histopathology under real-world adversarial examples"]}]}],"canonical_facts":{"dc:contributor":["Cong Shi","Lijing Wang","Xuan Liu"],"dc:creator":["Yajnik, Ruchik N."],"dc:date.created":["2026-05-31T07:00:00Z"],"dc:description.abstract":["<p>This robustness of histopathology classification models under adversarial and real-world perturbations resembling clinical artifacts is being investigated.</p><p>Using whole-slide images from the CAMELYON17 cohort, four representative architectures—ResNet-18, ResNet-50, HIPT-2MLP, and ViT-B/16 —are benchmarked across controlled pixel-level distortions and artifact-like transformations. Adversarial methods include iterative Fast Gradient Sign, Projected Gradient Descent, Salt-and-Pepper noise, and the Adversarial Watermark—Stain Shift (AWSS). Three defense strategies—Randomized Smoothing, Adversarial Training, and an Artifact Detector—are evaluated for their ability to preserve diagnostic accuracy and model reliability. Structured perturbations consistently degrade performance, with transformer-based models showing the greatest sensitivity. The benchmark developed here offers a reproducible framework for assessing robustness at both tile and slide levels, supporting safer deployment of histopathology AI in clinical workflows.</p>"],"dc:identifier":["https://digitalcommons.njit.edu/theses/3555"],"dc:subject":["Machine learning","Adversarial machine learning","Histopathology","Digital pathology","Medical imaging","Bioimaging and Biomedical Optics","Computer Sciences"],"dc:title":["Robustness of ai-driven histopathology under real-world adversarial examples"],"dc:type":["Thesis"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_name":["Master of Science in Computer Science - (M.S.)"]},"updated_at":"2026-11-19T08:44:11Z"}