TY - JOUR
T1 - Brain tumor segmentation using double density dual tree complex wavelet transform combined with convolutional neural network and genetic algorithm
AU - Samosir, Ridha Sefina
AU - Abdurachman, Edi
AU - Gaol, Ford Lumban
AU - Sabarguna, Boy Subirosa
N1 - Funding Information:
The authors would like to thank the Kalbis Institute of Technology and Business for funding this research. We would also like to thank Bina Nusantara University for providing the promoter and co-promoter for this research.
Publisher Copyright:
© 2022, Institute of Advanced Engineering and Science. All rights reserved.
PY - 2022/12
Y1 - 2022/12
N2 - Image segmentation is often faced by low contrast, bad boundaries, and inhomogeneity that made it difficult to separate normal and abnormal tissue. Therefore, it takes long periodto read and diagnose brain tumor patients. The aim of this study was to applied hybrid methods to optimize segmentation process of magnetic resonance image of brain. In this study, we divide the brain tumor images with double density dual-tree complex wavelet transform (DDDTCWT), continued by convolutional neural network (CNN), and optimized by genetic algorithm (GA) with 48 combinations yielding excellent results. The F-1 score was 99.42%, with 913 images test data. The training images consist of 1397 normal magnetic resonance imaging (MRI) images and 302 tumor MRI images resized by 32 x32 pixels. The DDDTCWT transforms the input images into more detail than ordinary wavelet transforms, and the CNNs will recognize the pattern of the output images. Additionally, we applied the GA to optimize the weights and biases from the first layer of the CNNs layers. The parameters used for evaluating were dice similarity coefficient (DSC), positive present value (PPV), sensitivity, and accuracy. The result showed that the combination of DDDTCWT, CNN, and GA could be used to brain MRI images and it generated parameters value more that 95%.
AB - Image segmentation is often faced by low contrast, bad boundaries, and inhomogeneity that made it difficult to separate normal and abnormal tissue. Therefore, it takes long periodto read and diagnose brain tumor patients. The aim of this study was to applied hybrid methods to optimize segmentation process of magnetic resonance image of brain. In this study, we divide the brain tumor images with double density dual-tree complex wavelet transform (DDDTCWT), continued by convolutional neural network (CNN), and optimized by genetic algorithm (GA) with 48 combinations yielding excellent results. The F-1 score was 99.42%, with 913 images test data. The training images consist of 1397 normal magnetic resonance imaging (MRI) images and 302 tumor MRI images resized by 32 x32 pixels. The DDDTCWT transforms the input images into more detail than ordinary wavelet transforms, and the CNNs will recognize the pattern of the output images. Additionally, we applied the GA to optimize the weights and biases from the first layer of the CNNs layers. The parameters used for evaluating were dice similarity coefficient (DSC), positive present value (PPV), sensitivity, and accuracy. The result showed that the combination of DDDTCWT, CNN, and GA could be used to brain MRI images and it generated parameters value more that 95%.
KW - Brain tumor
KW - Convolutional neural network
KW - Double density dual-tree complex wavelet transforms
KW - Genetic algorithm
KW - Image segmentation
UR - http://www.scopus.com/inward/record.url?scp=85136274189&partnerID=8YFLogxK
U2 - 10.11591/ijai.v11.i4.pp1373-1383
DO - 10.11591/ijai.v11.i4.pp1373-1383
M3 - Article
AN - SCOPUS:85136274189
SN - 2089-4872
VL - 11
SP - 1373
EP - 1383
JO - IAES International Journal of Artificial Intelligence
JF - IAES International Journal of Artificial Intelligence
IS - 4
ER -