TY - JOUR
T1 - Hybrid model of convolutional neural network and support vector machine to classify basal cell carcinoma
AU - Rojas, Jorge Alexander Ángeles
AU - Calderón Vilca, Hugo D.
AU - Tumi Figueroa, Ernesto N.
AU - Ramos, Kent Jhunior Cuadros
AU - Matos Manguinuri, Steve S.
AU - Calderón Vilca, Edwin F.
N1 - Publisher Copyright:
© 2021 Instituto Politecnico Nacional. All rights reserved.
PY - 2021
Y1 - 2021
N2 - Skin cancer is one of the most common types of cancer in humans, it covers about one third of all neoplasms. Within skin cancer we find basal cell carcinoma (BCC), this being the most frequent type of cancer worldwide. Solutions with convolutional neural networks generally use the Softmax layer (classic model) to perform a BCC classification, however, in other similar fields such as image classification of microscopic bacteria they have replaced this Softmax layer with a support vector machine (SVM) achieving a better result. Given this, we propose a hybrid model of convolutional neural network and a support vector machine (CNN+SVM) to classify the BCC. Our model is composed of 4 convolution blocks with 32, 64 and 128 filters to carry out the extraction of characteristics and then pass it to the classifier, to which the L1-SVM loss function is implemented. The average results obtained for the CNN+SVM hybrid model were measured with the precision, accuracy, recall and F1-score metrics, obtaining 96.200%, 96.200%, 96.205% and 96.200% respectively compared to the classical model for the metrics of precision, accuracy, recall and F1-score where 95.661%, 95.673%, 95.661%, 95.660% respectively were obtained. The results show that the hybrid model achieves better results than the classic model to classify the BCC.
AB - Skin cancer is one of the most common types of cancer in humans, it covers about one third of all neoplasms. Within skin cancer we find basal cell carcinoma (BCC), this being the most frequent type of cancer worldwide. Solutions with convolutional neural networks generally use the Softmax layer (classic model) to perform a BCC classification, however, in other similar fields such as image classification of microscopic bacteria they have replaced this Softmax layer with a support vector machine (SVM) achieving a better result. Given this, we propose a hybrid model of convolutional neural network and a support vector machine (CNN+SVM) to classify the BCC. Our model is composed of 4 convolution blocks with 32, 64 and 128 filters to carry out the extraction of characteristics and then pass it to the classifier, to which the L1-SVM loss function is implemented. The average results obtained for the CNN+SVM hybrid model were measured with the precision, accuracy, recall and F1-score metrics, obtaining 96.200%, 96.200%, 96.205% and 96.200% respectively compared to the classical model for the metrics of precision, accuracy, recall and F1-score where 95.661%, 95.673%, 95.661%, 95.660% respectively were obtained. The results show that the hybrid model achieves better results than the classic model to classify the BCC.
KW - Basal cell carcinoma
KW - Convolutional neural network
KW - Deep learning
KW - Support vector machine
UR - http://www.scopus.com/inward/record.url?scp=85102520087&partnerID=8YFLogxK
U2 - 10.13053/CYS-25-1-3431
DO - 10.13053/CYS-25-1-3431
M3 - Artículo
AN - SCOPUS:85102520087
SN - 1405-5546
VL - 25
SP - 83
EP - 95
JO - Computacion y Sistemas
JF - Computacion y Sistemas
IS - 1
ER -