RT Journal Article
T1 Ensemble-based genetic algorithm explainer with automized image segmentation: A case study on melanoma detection dataset
A1 Nematzadeh, Hossein
A1 García-Nieto, José Manuel
A1 Navas-Delgado, Ismael
A1 Aldana-Montes, José Francisco
K1 Inteligencia artificial - Aplicaciones médicas
K1 Melanoma
AB Explainable Artificial Intelligence (XAI) makes AI understandable to the human user particularly when themodel is complex and opaque. Local Interpretable Model-agnostic Explanations (LIME) has an image explainerpackage that is used to explain deep learning models. The image explainer of LIME needs some parameters tobe manually tuned by the expert in advance, including the number of top features to be seen and the numberof superpixels in the segmented input image. This parameter tuning is a time-consuming task. Hence, with theaim of developing an image explainer that automizes image segmentation, this paper proposes Ensemblebased Genetic Algorithm Explainer (EGAE) for melanoma cancer detection that automatically detects andpresents the informative sections of the image to the user. EGAE has three phases. First, the sparsity ofchromosomes in GAs is determined heuristically. Then, multiple GAs are executed consecutively. However,the difference between these GAs are in different number of superpixels in the input image that result indifferent chromosome lengths. Finally, the results of GAs are ensembled using consensus and majority votings.This paper also introduces how Euclidean distance can be used to calculate the distance between the actualexplanation (delineated by experts) and the calculated explanation (computed by the explainer) for accuracymeasurement. Experimental results on a melanoma dataset show that EGAE automatically detects informativelesions, and it also improves the accuracy of explanation in comparison with LIME efficiently. The pythoncodes for EGAE, the ground truths delineated by clinicians, and the melanoma detection dataset are availableat https://github.com/KhaosResearch/EGAE
PB Elsevier
YR 2023
FD 2023
LK https://hdl.handle.net/10630/26388
UL https://hdl.handle.net/10630/26388
LA eng
NO Nematzadeh, García-Nieto, J., Navas-Delgado, I., & Aldana-Montes, J. F. (2023). Ensemble-based genetic algorithm explainer with automized image segmentation: A case study on melanoma detection dataset. Computers in Biology and Medicine, 155, 106613–106613. https://doi.org/10.1016/j.compbiomed.2023.106613
NO This work has been partially funded by grant PID2020-112540RBC41 (funded by MCIN/AEI/10.13039/501100011033/, Spain), AETHERUMA, Spain (A smart data holistic approach for context-aware data analytics: semantics and context exploitation). Funding for open access charge: Universidad de Málaga/CBUA. Additionally, we thank Dr. Miguel Ángel Berciano Guerrero from Unidad de Oncología Intercentros, Hospitales Univesitarios Regional 𝑦 Virgen de la Victoria de Málaga, and Instituto de Investigaciones Biomédicas (IBIMA), Málaga, Spain, for his support in images selection and general medical orientation in the particular case of Melanoma.
DS RIUMA. Repositorio Institucional de la Universidad de Málaga
RD 20 ene 2026