Artificial Intelligence Based Melanoma Early Diagnosis and Risk Prediction in Children, Adolescents and Young Adults

German Cancer Research Center3,000 enrolled

Overview

The goal of this study is to develop supportive diagnostic artificial intelligence algorithms to distinguish melanoma from nevi or other benign pigmented skin lesions, especially in younger patients (below the age of 30). The main goals it aims to achieve are: * development of an algorithm based on dermatoscopic images, targeting skin cancer screening in vulnerable populations * development of another algorithm based on histological images, intended to be used by pathologists on lesions that are still suspicious of melanoma after dermatologic assessment * implementation of explainability methods to enable the user to better comprehend the systems' decisions, avoid biases and increase trust in these applications There is no additional time commitment for the study participants for this study, as the data used in this project will be collected in routine clinical practice anyway.

Study Type

OBSERVATIONAL

Enrollment

3,000

Conditions

Melanoma (Skin Cancer)Artificial Intelligence (AI)Pediatric Cancer

Eligibility

Sex: ALLHealthy volunteers:

Medical Language ↔ Plain English

Inclusion Criteria: \- Exclusion Criteria: * Patients without a melanoma or nevus diagnosis * images with insufficient image quality

Locations (3)

University of Tübingen

Tübingen, Germany

COMPLETED

University of Florence

Florence, Italy

COMPLETED

Hospital Clínic de Barcelona

Barcelona, Spain

RECRUITING

Outcomes

Primary Outcomes

Area Under the Receiver Operator Curve (AUROC)

The AUROC is used to measure and compare the diagnostic accuracy of different classifiers. Thereby, a higher value means better diagnostic performance, with an AUROC of 1 being a perfect score.

Time frame: First Assessment: Upon completion of the first training and testing cycle (approx. within 1.5 years from the start of the study). Reevaluations: at 6 and 12 months post-initial training for model improvement.

Secondary Outcomes

Balanced accuracy

The balanced accuracy is used to measure and compare the diagnostic accuracy between classifier and physician. Thereby, a higher value means better diagnostic performance, with a balanced accuracy of 1 signifying perfect diagnostic capabilities.