Por favor, use este identificador para citar o enlazar este ítem: https://repositorio.cetys.mx/handle/60000/1750
Título : Multi-Class diagnosis of skin lesions using the fourier spectral information of images on additive color model by Artificial Neural Network
Otros títulos : IEEE Access
Autor : López-Leyva, Josué Aarón
Guerra-Rosas, Esperanza
Álvarez-Borrego, Josué
Palabras clave : Artificial neural networks;Biomedical computing;Image processing;Medical diagnostic imaging;Fourier spectral analysis
Sede: Campus Ensenada
Fecha de publicación : 24-abr-2021
Citación : vol. 9;
Resumen : This article presents a new methodology to diagnostics ten types of skin lesions based on the image’ s Fourier spectral information in an additive color model. All spectral information and correlation coefficients between the skin lesions classes conform the input signals to an Artificial Neural Network. In general, the results show the well-defined classification for all the skin lesions classes based on the high values for Accuracy, Precision, Sensitivity, and Specificity metrics performance and a reduced images misclassification percentage (≈5.9%) for the Testing sub-dataset, and less for Training (≈2.8%) and Validation (≈5.6%) sub-dataset even considering the strange objects, not-clarity, and black sections in some images analyzed. The general achieved classification Accuracy, Precision, Sensitivity, and Specificity percentages of the proposed method are 99.33 %, 94.16 %, 92.9 %, and 99.63 %, respectively. In particular, the skin lesions related to Basal Cell Carcinoma, Seborrhoeic Keratosis, and Melanocytic Nevus present the best performance regarding the Receiver Operating Characteristics, while the Pyogenic Granuloma was the worst classified
metadata.dc.description.url: https://ieeexplore.ieee.org/document/9363122
URI : https://repositorio.cetys.mx/handle/60000/1750
Aparece en las colecciones: Artículos de Revistas

Este ítem está protegido por copyright original

Este ítem está sujeto a una licencia Creative Commons Licencia Creative Commons Creative Commons