Multi‐level dilated convolutional neural network for brain tumour segmentation and multi‐view‐based radiomics for overall survival prediction

Abstract

Glioblastoma (GBM) is the most high-risk and grievous tumour in the brain that causes the death of more than 50% of the patients within one to 2 years after diagnosis. Accurate detection and prognosis of this disease are critical to provide essential guidelines for treatment planning. This study proposed using a deep learning-based network for the GBM segmentation and radiomic features for the patient's overall survival (OS) time prediction. The segmentation model used in this study was a modified U-Net-based deep 3D multi-level dilated convolutional neural network. It uses multiple kernels of altered sizes to capture contextual information at different levels. The proposed scheme for OS time prediction overcomes the problem of information loss caused by the derivation of features in a single view due to the variation in the neighbouring pixels of the tumorous region. The selected features were based on texture, shape, and volume. These features were computed from the segmented components of tumour in axial, coronal, and sagittal views of magnetic resonance imaging slices. The proposed models were trained and evaluated on the BraTS 2019 dataset. Experimental results of OS time prediction on the validation data have showed an accuracy of 48.3%, with the mean squared error of 92 599.598. On the validation data, the segmentation model achieved a mean dice similarity coefficient of 0.75, 0.89, and 0.80 for enhancing tumour, whole tumour, and tumour core, respectively. Future work is warranted to improve the overall performance of OS time prediction based on the findings in this study.

Brain tumor segmentation and multiview multiscale‐based radiomic model for patient's overall survival prediction

“…For the performance evaluation of the segmentation model, four metrics named DSC, sensitivity, specificity, and the HD are used. 7 It can be seen that this study has outperformed the existing techniques 10,16,17,30,34,48 for brain tumor segmentation. By analyzing the BraTS2019 and BraTS2020 datasets, the BraTS2020 dataset outperformed due to its larger training set.…”

“…This study used five MLDC blocks for tumor segmentation with different dilation rates, shown in Figure 2. MLDCNN-123, MLDCNN-135, MLDC-357, MLDC-246, and MLDC-468 are implemented by setting the dilation rates of (1, 2, 3); (1,3,5); (3,5,7); (2,4,6); and (4,6,8), respectively, at each level of MLDC block. In the MLDCNN model, standard convolution is replaced with the 3D dilated convolution.…”

“…Our previous study proposed a multilevel dilated convolution neural network (MLDCNN) for brain tumor segmentation. 7 Dilated convolutions were used to enlarge the receptive field with a lesser number of parameters than standard convolution. Moreover, dilation at three different levels was performed to incorporate the features at multiple scales (fine, medium, and coarse) and mitigate the skipping pixel problem of dilated convolutions.…”

“…For the analysis of image-based on multiscale texture, LoG and wavelet-based methods have been incorporated. 1,7,18,23 However, in existing literature, LoG-based multiscale texture features were extracted using narrower Gaussian, which are more susceptible to noise in the input image; furthermore, features extracted using wider Gaussian were found more predictive than features obtained at narrower Gaussian. 20 For wavelet-based multiscale texture features, 2D wavelet transform is used to decompose the region of interest (ROI) into horizontal and vertical dimensions for OS prediction.…”

“…4 Unlike the existing literature, extensive experimentation has been performed, in this study, for the comparative analysis of different dilation rates in the brain tumor segmentation task. Subsequently, this study has implemented the five variations of the MLDCNN model for brain tumor segmentation with varying rates of dilation in the MLDC block at three levels such as (1, 2, 3); (1,3,5); (3,5,7); (2,4,6); and (4,6,8). In this way, each MLDCNN model extracts local and global features at different scales.…”