Joint training of two-channel deep neural network for brain tumor classification
- PDF / 1,150,185 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 84 Downloads / 178 Views
ORIGINALPAPER
Joint training of two-channel deep neural network for brain tumor classification Jyostna Devi Bodapati1
· Nagur Shareef Shaik1 · Veeranjaneyulu Naralasetti1 · Nirupama Bhat Mundukur1
Received: 15 July 2020 / Revised: 13 September 2020 / Accepted: 22 September 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Brain tumor recognition is a challenging task, and accurate diagnosis increases the chance of patient survival. In this article, we propose a two-channel deep neural network architecture for tumor classification that is more generalizable. Initially, local feature representations are extracted from convolution blocks of InceptionResNetV2 and Xception networks and are vectorized using proposed pooling-based techniques. An attention mechanism is proposed that allows more focus on tumor regions and less focus on non-tumor regions which eventually helps to differentiate the type of tumor present in the images. The proposed two-channel model allows joint training of two sets of tumor image representations in an end-to-end manner to achieve good generalization. Empirical studies on Figshare and BraT’S2018, benchmark datasets, reveal that our approach is superior in terms of generalization and simple in terms of number of layers compared to the existing complex models that follow fine-tuning of deep CNN models. Avoiding too much preprocessing and augmentation techniques, the proposed model sets new state-of-the-art scores on both the brain tumor datasets. Keywords Deep learning · Brain tumor classification (BTC) · Pre-trained convolution neural networks (CNNs) · InceptionResNetV2 · Xception · Pooling-based vectorization · Attention mechanism · Two-channel deep neural network (DNN)
1 Introduction Brain tumor is a deadly disease caused by an uncontrolled and abnormal growth of unwanted cells creating a mass of tissues in the brain affecting the central nervous system [17]. Manual diagnosis of brain tumor involves taking and screening magnetic resonance (MR) imaging scans of brain to detect the presence of tumor [5]. In recent years, plenty of machine learning-based computer-aided diagnostic (CAD) tools have been in use aiding radiologists. The initial models used for tumor prediction were trivial and use feature representations like wavelet features and machine learning models for detecting whether the tumor is normal or abnormal [19]. The next class of models uses multiple sets (texture, boundary and shape) of features together and these highdimensional features are reduced by feature selection and reduction methods to select significant features. This reduced
B 1
Jyostna Devi Bodapati [email protected] Department of Computer Science and Engineering, Vignan’s Foundation for Science Technology and Research, Guntur 522213, India
feature set is trained on ensemble of classifiers that include knearest neighbor (k-NN), SVM and artificial neural network (ANN) [4]. Khalid et al. utilized intensity and neighborhood along with wavelet features of brain MR images. Such features ar
Data Loading...
