Methodological Comparison of Convolutional Neural Networks and Kolmogorov-Arnold Networks in Melanoma Detection
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Computer Science and Information Systems
Date of Award
Spring 2025
Abstract
Detecting melanoma, a deadly skin cancer, is crucial for improving patient outcomes. This study compares two popular contemporary methods of Convolutional Neural Networks (CNNs) and Kolmogorov-Arnold Networks (KANs) for melanoma detection. CNNs, widely used in image analysis, can automatically learn and extract features from large datasets. KANs are a newer architecture designed to enhance the ability of CNNs to model long-range dependencies and complex patterns in images. They incorporate attention mechanisms directly into the kernel operations used in convolutional layers. This study compares the accuracy, computational time, and network structure of these models using a standardized melanoma dataset, a subset of large database supported by ISIC consortium. It is hypothesized that newly introduced KANs model is trained faster with low number of neurons to obtain similar results of CNN models. Experiment result showed that KAN cannot outperform in many skin lesion classification tasks, having similar or slightly worse accuracies. in addition to melanoma detection, we would also like to evaluate both algorithms and compare them in general classification and recognition problems. Therefore, we compared both algorithms in flower recognition, a different problem. This research contributes to the development of automated melanoma detection systems and provide insights into the trade-offs between different neural network architectures for medical image analysis. Future studies will focus on comparison on other medical image domains, such as ultrasound and radiology images of other diseases.
Advisor
Kaoning Hu
Subject Categories
Computer Sciences | Physical Sciences and Mathematics
Recommended Citation
Mete, Bayram Bahadir, "Methodological Comparison of Convolutional Neural Networks and Kolmogorov-Arnold Networks in Melanoma Detection" (2025). Electronic Theses & Dissertations. 1267.
https://digitalcommons.tamuc.edu/etd/1267
