{"id":{"repo_id":"calpoly","oai_identifier":"oai:digitalcommons.calpoly.edu:theses-4926"},"canonical_url":"https://search.ndltd.org/etd/calpoly/oai:digitalcommons.calpoly.edu:theses-4926","repository":{"repo_id":"calpoly","name":"Cal Poly","base_url":"https://digitalcommons.calpoly.edu/do/oai/"},"display":{"title":"Skinclusive AI: Towards Equitable Skin Cancer Detection for Deployment on Edge Devices","abstract":"Skin cancer is one of the most prevalent cancers worldwide, yet existing deep-learning models exhibit significant racial disparities because many widely used datasets are heavily skewed toward lighter skin tones. In addition, many approaches are not designed for deployment on resource-constrained devices, which limits accessibility. This work presents a comprehensive evaluation of classical machine learning and deep-learning based models for binary skin lesion classification, identifying the Swin-Tiny transformer architecture as the most effective backbone. To address bias, we curate a skin-tone balanced dataset, and introduce fairness-aware training through adversarial training, and joint distribution oversampling, to improve performance across protected attributes. The final model is converted to ONNX format and quantized for deployment on both general-purpose and resource-constrained computer hardware, including an Intel Xeon (via Google Colab) and a Raspberry Pi 3 (RPi3). Our final quantized system achieved an average inference of 1.91 seconds per image on a RPi3. Additionally, we achieve a precision of 0.867, recall of 0.891, and an F1-score of 0.879, with a PR-AUC of 0.928 while reducing disparities across skin tones, achieving an equal opportunity difference of 0.095 and an average odds difference of 0.148 across demographic groups. These results demonstrate the feasibility of combining fairness-aware learning with efficient edge deployment to support more accessible and equitable skin cancer screening.","abstract_html":"Skin cancer is one of the most prevalent cancers worldwide, yet existing deep-learning models exhibit significant racial disparities because many widely used datasets are heavily skewed toward lighter skin tones. In addition, many approaches are not designed for deployment on resource-constrained devices, which limits accessibility. This work presents a comprehensive evaluation of classical machine learning and deep-learning based models for binary skin lesion classification, identifying the Swin-Tiny transformer architecture as the most effective backbone. To address bias, we curate a skin-tone balanced dataset, and introduce fairness-aware training through adversarial training, and joint distribution oversampling, to improve performance across protected attributes. The final model is converted to ONNX format and quantized for deployment on both general-purpose and resource-constrained computer hardware, including an Intel Xeon (via Google Colab) and a Raspberry Pi 3 (RPi3). Our final quantized system achieved an average inference of 1.91 seconds per image on a RPi3. Additionally, we achieve a precision of 0.867, recall of 0.891, and an F1-score of 0.879, with a PR-AUC of 0.928 while reducing disparities across skin tones, achieving an equal opportunity difference of 0.095 and an average odds difference of 0.148 across demographic groups. These results demonstrate the feasibility of combining fairness-aware learning with efficient edge deployment to support more accessible and equitable skin cancer screening.","abstract_has_math":false,"creators":["Galicinao, Joseph"],"institution":null,"degree_name":"MS in Electrical Engineering","degree_level":null,"degree_discipline":"Electrical Engineering","degree_department":null,"school":null,"contributors":["Jane Zhang","Electrical Engineering","College of Engineering"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T07:12:51Z","subjects":["Skin cancer detection","deep learning","fairness in machine learning","edge computing","Electrical and Computer Engineering"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.calpoly.edu/theses/3261","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.calpoly.edu/theses/3261","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.calpoly.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.calpoly.edu%3Atheses-4926","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Jane Zhang","Electrical Engineering","College of Engineering"]},{"key":"dc:creator","label":"Author","values":["Galicinao, Joseph"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2026-05-27T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2026-06-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering"]},{"key":"thesis:degree_name","label":"Degree Name","values":["MS in Electrical Engineering"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Skin cancer detection","deep learning","fairness in machine learning","edge computing","Electrical and Computer Engineering"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.calpoly.edu/theses/3261"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Skin cancer is one of the most prevalent cancers worldwide, yet existing deep-learning models exhibit significant racial disparities because many widely used datasets are heavily skewed toward lighter skin tones. In addition, many approaches are not designed for deployment on resource-constrained devices, which limits accessibility. This work presents a comprehensive evaluation of classical machine learning and deep-learning based models for binary skin lesion classification, identifying the Swin-Tiny transformer architecture as the most effective backbone. To address bias, we curate a skin-tone balanced dataset, and introduce fairness-aware training through adversarial training, and joint distribution oversampling, to improve performance across protected attributes. The final model is converted to ONNX format and quantized for deployment on both general-purpose and resource-constrained computer hardware, including an Intel Xeon (via Google Colab) and a Raspberry Pi 3 (RPi3). Our final quantized system achieved an average inference of 1.91 seconds per image on a RPi3. Additionally, we achieve a precision of 0.867, recall of 0.891, and an F1-score of 0.879, with a PR-AUC of 0.928 while reducing disparities across skin tones, achieving an equal opportunity difference of 0.095 and an average odds difference of 0.148 across demographic groups. These results demonstrate the feasibility of combining fairness-aware learning with efficient edge deployment to support more accessible and equitable skin cancer screening."]},{"key":"dc:title","label":"Title","values":["Skinclusive AI: Towards Equitable Skin Cancer Detection for Deployment on Edge Devices"]}]}],"canonical_facts":{"dc:contributor":["Jane Zhang","Electrical Engineering","College of Engineering"],"dc:creator":["Galicinao, Joseph"],"dc:date.available":["2026-05-27T07:00:00Z"],"dc:date.created":["2026-06-01T07:00:00Z"],"dc:description.abstract":["Skin cancer is one of the most prevalent cancers worldwide, yet existing deep-learning models exhibit significant racial disparities because many widely used datasets are heavily skewed toward lighter skin tones. In addition, many approaches are not designed for deployment on resource-constrained devices, which limits accessibility. This work presents a comprehensive evaluation of classical machine learning and deep-learning based models for binary skin lesion classification, identifying the Swin-Tiny transformer architecture as the most effective backbone. To address bias, we curate a skin-tone balanced dataset, and introduce fairness-aware training through adversarial training, and joint distribution oversampling, to improve performance across protected attributes. The final model is converted to ONNX format and quantized for deployment on both general-purpose and resource-constrained computer hardware, including an Intel Xeon (via Google Colab) and a Raspberry Pi 3 (RPi3). Our final quantized system achieved an average inference of 1.91 seconds per image on a RPi3. Additionally, we achieve a precision of 0.867, recall of 0.891, and an F1-score of 0.879, with a PR-AUC of 0.928 while reducing disparities across skin tones, achieving an equal opportunity difference of 0.095 and an average odds difference of 0.148 across demographic groups. These results demonstrate the feasibility of combining fairness-aware learning with efficient edge deployment to support more accessible and equitable skin cancer screening."],"dc:identifier":["https://digitalcommons.calpoly.edu/theses/3261"],"dc:subject":["Skin cancer detection","deep learning","fairness in machine learning","edge computing","Electrical and Computer Engineering"],"dc:title":["Skinclusive AI: Towards Equitable Skin Cancer Detection for Deployment on Edge Devices"],"thesis:degree_discipline":["Electrical Engineering"],"thesis:degree_name":["MS in Electrical Engineering"]},"updated_at":"2026-11-19T07:12:51Z"}