TransparentAI: High-Accuracy and Explainable Melanoma Classification System
Principal Investigator
Name
Björn Friedrich
Degrees
Ph.D.
Institution
Carl von Ossietzky University, Medicine and Health Sciences, Department of Health Services Research, Division for Assistance Systems and Medical Device Technology
Position Title
Associated Senior Researcher
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-1623
Initial CDAS Request Approval
Jul 15, 2024
Title
TransparentAI: High-Accuracy and Explainable Melanoma Classification System
Summary
The TransparentAI project aims to develop a high-accuracy melanoma classification system, prioritizing explainability and compliance with legal regulations. By combining data from diverse sources, we create a robust and comprehensive dataset to enhance the algorithm's reliability. While performance metrics such as AUC and ROC are important, our primary focus is on ensuring the system's transparency and interpretability.
We adopt a user-centric approach, engaging in interdisciplinary research to understand and address the needs of dermatologists. By collaborating closely with medical professionals, we ensure that the system provides explanations they can trust and rely on for making informed decisions. This approach not only fosters trust in the algorithm but also ensures its practical utility in clinical settings.
Aims
- How can we integrate and optimize diverse data sources to create a comprehensive and high-quality dataset for melanoma classification?
This question addresses the challenge of aggregating and harmonizing data from various origins to enhance the model's training and performance.
-What methods can be employed to ensure the explainability and transparency of the melanoma classification algorithm, making it understandable and trustworthy for dermatologists?
This focuses on identifying techniques and frameworks that improve the interpretability of the model, ensuring that its decision-making process is clear and justifiable to medical professionals.
- How can we balance high-performance metrics (AUC and ROC) with the need for explainability and legal compliance in the development of the melanoma classification system?
This question explores the trade-offs and strategies for achieving optimal performance while adhering to regulatory requirements and maintaining the system's explainability.
Collaborators
Dr. Björn Friedrich, Carl von Ossietzky University, Oldenburg, Germany
Daniel Eckhoff, Ph.D., Huawei Research, Hongkong, Hongkong SAR
Prof. Dr. med. Ulrike Raap, University Hospital, Oldenburg, Germany
Prof. Dr.-Ing. Andreas Hein, Carl von Ossietzky University, Oldenburg, Germany