Publicação
Optimizing melanoma screening: AI-based risk stratification for efficient patient prioritization
| dc.contributor.author | Silva, Margarida | |
| dc.contributor.author | Prudêncio, Cristina | |
| dc.contributor.author | Coelho, Pedro | |
| dc.contributor.author | Coelho, Pedro | |
| dc.date.accessioned | 2026-07-28T09:13:01Z | |
| dc.date.available | 2026-07-28T09:13:01Z | |
| dc.date.issued | 2026-05-29 | |
| dc.description.abstract | Melanoma is a malignant neoplasm derived from melanocytes that is highly curable when detected early. However the 5-year survival rate drops from 99% to 35% when diagnosed at advanced stages [1,2]. In Portugal, constraints in the public health care system (SNS) waiting lists may compromise this critical treatment window [3]. This word presents the development of an artificial intel-ligence (AI)-powered clinical decision support system designed to assist dermatologists by pre-classifying melanoma risk in dermoscopic images obtained in primary healthcare settings, while optimizing patient prioritization. The proposed image classification methodology employs a two-stage deep learning computer vision pipeline implemented in Python using TensorFlow and trained on the publicly available ISIC 2020 and HAM10000 skin lesion datasets. First, skin lesions are identified, isolated, and segmented using a U-Net architecture. Subsequently, the seg-mented lesion masks are classified using an Efficient-Net-based model to estimate malignancy probability. The trained models were integrated into a real-time web application developed using the Reflex frame-work. Preliminary results demonstrate that the U-Net architecture effectively segments lesions, reducing the influence of irrelevant artifacts during classification. Following segmentation, the Efficient-Net classifier achieved an accuracy of 98.67% and an area under the receiver operating characteristic curve (AUC) of 99.8% on a test set comprising 4,500 der-moscopic images. Importantly for clinical screening applications, the model achieved a sensitivity (recall) of 0.99 for malignant lesions, corresponding to only 17 false negatives among 2,250 malignant cases. The proposed two-stage deep-learning pipeline shows potential as a clinical support tool for dermatologists. By providing automated and interpre-table AI-based risk assessments, the system may as-sist physicians in prioritizing urgent cases with greater confidence. As a proof-of-concept, this tool has the potential to augment clinical workflows and help re-duce diagnostic delays, particularly within the public healthcare system. | eng |
| dc.identifier.citation | Silva, M., Prudêncio, C., & Coelho, P. (2026). Optimizing melanoma screening: AI-based risk stratification for efficient patient prioritization. Book of Abstracts of the 8th Meeting on Medicinal Biotechnology, 58. https://edicoes.ipp.pt/index.php/books/catalog/book/251 | |
| dc.identifier.doi | 10.26537/ed.p.porto.251 | |
| dc.identifier.isbn | 978-989-9226-20-3 | |
| dc.identifier.uri | http://hdl.handle.net/10400.22/32614 | |
| dc.language.iso | eng | |
| dc.peerreviewed | yes | |
| dc.publisher | Politema | |
| dc.relation.hasversion | https://edicoes.ipp.pt/index.php/books/catalog/book/251 | |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc-nd/4.0/ | |
| dc.subject | Melanoma | |
| dc.subject | Deep learning | |
| dc.subject | Dermatology | |
| dc.subject | Triage | |
| dc.title | Optimizing melanoma screening: AI-based risk stratification for efficient patient prioritization | eng |
| dc.type | conference object | |
| dspace.entity.type | Publication | |
| oaire.citation.conferenceDate | 2026-05-29 | |
| oaire.citation.conferencePlace | Porto | |
| oaire.citation.endPage | 58 | |
| oaire.citation.startPage | 58 | |
| oaire.citation.title | Book of Abstracts of the 8th Meeting on Medicinal Biotechnology | |
| oaire.version | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |
| person.familyName | Coelho | |
| person.givenName | Pedro | |
| person.identifier.ciencia-id | BE18-3DB2-8D79 | |
| person.identifier.orcid | 0000-0002-2343-1108 | |
| person.identifier.rid | AAZ-9239-2020 | |
| person.identifier.scopus-author-id | 57214054449 | |
| relation.isAuthorOfPublication | eb030614-4140-437b-8ac3-f59c52566ecf | |
| relation.isAuthorOfPublication.latestForDiscovery | eb030614-4140-437b-8ac3-f59c52566ecf |