Brain tumor detection: a long short-term memory (LSTM)-based learning model
- PDF / 1,225,105 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 5 Downloads / 218 Views
(0123456789().,-volV)(0123456789(). ,- volV)
RECENT ADVANCES IN DEEP LEARNING FOR MEDICAL IMAGE PROCESSING
Brain tumor detection: a long short-term memory (LSTM)-based learning model Javaria Amin1 • Muhammad Sharif1
•
Mudassar Raza1 • Tanzila Saba2 • Rafiq Sial3 • Shafqat Ali Shad4
Received: 15 December 2018 / Accepted: 22 November 2019 Ó Springer-Verlag London Ltd., part of Springer Nature 2019
Abstract To overcome the problems of automated brain tumor classification, a novel approach is proposed based on long short-term memory (LSTM) model using magnetic resonance images (MRI). First, N4ITK and Gaussian filters having size 5 9 5 are used to boost the of multi-sequence MRI quality. The presented deep LSTM model having four layers is utilized for classification. In each layer, optimal hidden units (HU) are selected such as 200 HU, 225 HU, 200 HU and 225 HU, respectively. These hidden or concealed units are chosen after performing extensive experiments to acquire better results. The results are validated on different versions of BRATS datasets (BRATS 2012–15, 2018) and SISS-ISLES 2015 dataset. The presented method attained dice similarity coefficient (DSC) 1.00 on 2012 synthetic, 0.95 on 2013, 0.99 on 2013 Leader board, 0.99 on 2014, 0.98 on 2015, 0.99 on 2018 and 0.95 on SISS-ISLES 2015. The methodology is also checked on real patient’s cases of brain tumor collected from Pakistan ordinance factory and achieved 0.97 DSC. The results confirm that the presented method provides more help for radiologists to classify brain tumor precisely. Keywords MRI LSTM HU Brain tumor Detection Prediction List xt ht Ot B ct R R
of symbols Input image Hidden layer Model output Bias Cell state time step Recurrent weights Sigmoid activation function Hadamard product
& Muhammad Sharif [email protected] 1
Department of Computer Science, COMSATS University Islamabad, Wah Campus, Islamabad, Pakistan
2
College of Computer and Information Sciences, Prince Sultan University, Riyadh, Saudi Arabia
3
Department of Mathematics, COMSATS University Islamabad, Wah Campus, Islamabad, Pakistan
4
Department of Computer Science, Luther College, Decorah, IA, USA
Abbreviations RNN Recurrent neural network NNs Feedforward neural networks SE Sensitivity SP Specificity FN False negative FP False positive TN True negative TP True positive DSC Dice similarity coefficient DWI Diffusion-weighted imaging FNR False negative rate FLAIR Fluid-attenuated inversion recovery T1c T1-weighted contrast-enhanced T1 T1-weighted RF Random forests SVMs Support vector machines CNNs Convolutional neural networks MRFs Markov random fields CEN Convolutional encoder networks HGG High-grade glioma CRFs Conditional random fields LGG Low-grade glioma KNN K-nearest neighbor DT Decision trees
123
Neural Computing and Applications
MRI JSI FPR PPV ACC
Magnetic resonance images Jaccard similarity index False positive rate Positive predictive value Accuracy
1 Introduction Irregular cells development [1–5] is the main cause of brain tumor. Glioma is an
Data Loading...
