Explainable Artificial Intelligence for Detection of Structural Changes in Myocardium
Loading...
Files
Date
Publisher
Philipps-Universität Marburg
Supervisors
Abstract
Artificial Intelligence (AI) has achieved significant success in various fields, including medicine. However, AI systems are often perceived as "black boxes" due to their opaque nature, raising concerns about their reliability and trustworthiness. Explainable AI (XAI) seeks to address these concerns by enhancing the transparency of AI systems and thereby making them more accessible to humans. Despite its potential, selecting appropriate XAI methods for specific applications remains a challenge, largely due to a lack of standardized evaluation criteria. This thesis focuses on the comparison and evaluation of XAI methods to improve trust and to encourage the adoption of AI in medical practice, particularly in the context of cardiovascular diseases, which are one of the leading causes of death globally. Through an end-to-end use case, this thesis illustrates how AI systems can support clinical decision-making and further explores the role of XAI in increasing the acceptance of this technology among medical professionals. Initial experiments revealed biases in existing XAI methods, leading to the development of a novel approach called SIGN (Sign-based Improvement of Gradient-based explaNations). Comparative evaluations on benchmark datasets demonstrated the superiority of SIGN over state-of-the-art XAI methods in the domain of computer vision. Furthermore, a subsequent study validated the reliability and medical interpretability of SIGN-based explanations for electrocardiogram time series data, aligning with diagnostic patterns in established medical guidelines. These findings highlight the potential of SIGN for enhancing the trustworthiness of AI systems in cardiology, contributing to more reliable and interpretable clinical decision support tools.
Review
Metadata
Contributors
Supervisor:
Dates
Created: 2024Issued: 2025-06-26Updated: 2025-06-26
Faculty
Fachbereich Mathematik und Informatik
Publisher
Philipps-Universität Marburg
Language
eng
Data types
DoctoralThesis
Keywords
Guideline PatternsEKGTrustExplainable Artificial IntelligenceElectrocardiogramDeep LearningArtificial IntelligenceInformation SciencesSIGNPattern RecognitionCardiology
DDC-Numbers
004
show more
Gumpfer, Nils (0000-0001-8644-9885): Explainable Artificial Intelligence for Detection of Structural Changes in Myocardium. : Philipps-Universität Marburg 2025-06-26. DOI: https://doi.org/10.17192/z2025.0472.