3D multi-scale residual fully convolutional neural network for segmentation of extremely large-sized kidney tumor. (March 2022)

• Record Type: Journal Article
• Authors: Yang, Ehwa; Kim, Chan Kyo; Guan, Yi; Koo, Bang-Bon; Kim, Jae-Hun
• Is Part Of: Computer methods and programs in biomedicine. Volume 215(2022)
• Publication Date: 2022-03
• Subjects: MRI Magnetic Resonance Imaging -- CT Computerized Tomography -- PET Positron Emission Tomograpy -- CNN Convolutional Neural Network -- FCN Fully Convolutional Neural Network -- Res3D U-Net 3D ResNet with U-Net -- 3D-MS-RFCN 3D Multi-Scale Residual Fully Convolutional Neural Network -- KiTS Kidney Tumor Segmentation Challenge -- PA Pixel Accuracy -- HD Hausdorff Distance -- DICE Dice Similarity Coefficeint; Kidney -- Kidney tumor -- Medical image -- Segmentation -- Deep learning -- Fully convolutional neural network; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2022.106616
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: From feature map analysis using Res3D U-Net model, we found the under-segmentation of the extremely large sized kidney tumors in CT image due to the lack of the global contextual information. To mitigate this problem, we proposed 3D-MS-RFCNN model to improve the extremely large sized kidney tumors by adding an additional encoding path for extracting the global contextual information. Our sub-group analysis showed the improvement of our 3D-MS-RFCNN model for segmenting the extremely large sized tumors. Feature map analysis using 3D-MS-RFCNN model confirmed the effect of an additional encoding path for extracting the global contextual information from the extremely large sized kidney tumors. Ensemble model using Res3D U-Net and 3D-MS-RFCNN models for balanced performance for segmenting the various size of kidney tumors. Abstract: Background and objective: We propose a novel deep neural network, the 3D Multi-Scale Residual Fully Convolutional Neural Network (3D-MS-RFCNN) to improve segmentation in extremely large-sized kidney tumors. Method: The multi-scale approach with a deep neural network is applied to capture global contextual features. Our method, 3D-MS-RFCNN, consists of two encoders and one decoder as a single complete network. One of the encoders is designed for capturing global contextual information by using the low-resolution, down-sampled data from input images. In the decoder, features from the encoder for global contextual features are concatenated with up-sampled features from the previous layer and features from the other encoder. Ensemble learning strategy is also applied. Results: We evaluated the performance of our proposed method using the KiTS public dataset and the in-house hospital dataset. When compared with the state-of-the-art method, Res3D U-Net, our model, 3D-MS-RFCNN, demonstrated greater accuracy (0.9390 dice score for KiTS dataset and 0.8575 dice score for external dataset) for segmenting extremely large-sized kidney tumors. Conclusions: Our proposed network shows significantly improved segmentation performance of extremely large-sized targets. This study can be usefully employed in the field of medical image analysis.