Volume segmentation using convolutional neural networks with limited training data

Much of the success of convolutional neural networks (CNNs) is due to the enormous collections of labeled data, powerful GPUs, and modern network architectures that facilitate the training and testing of larger and deeper models. The limited size of labeled volumetric microscopy images, however, hinders the proper training of CNNs for volume segmentation. Here, we design various 2D and 3D CNNs with factorized convolutions and online feature-level augmentations to address challenges due to the scarcity of training data. Based on the experimental results, we found that the 3D CNNs consistently outperformed the 2D counterparts. In addition, the 3D CNN that uses both factorized convolutions and online feature-level augmentations achieved the best segmentation performance.

[1]  Pascal Fua,et al.  Learning Structured Models for Segmentation of 2-D and 3-D Imagery , 2015, IEEE Transactions on Medical Imaging.

[2]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[3]  Junmo Kim,et al.  Deep Pyramidal Residual Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Lorenzo Torresani,et al.  Learning Spatiotemporal Features with 3D Convolutional Networks , 2014, 2015 IEEE International Conference on Computer Vision (ICCV).

[5]  Amitabh Varshney,et al.  Visualization of Brain Microstructure Through Spherical Harmonics Illumination of High Fidelity Spatio-Angular Fields , 2014, IEEE Transactions on Visualization and Computer Graphics.

[6]  Jian Sun,et al.  Identity Mappings in Deep Residual Networks , 2016, ECCV.

[7]  Joseph JáJá,et al.  Hierarchical Exploration of Volumes Using Multilevel Segmentation of the Intensity-Gradient Histograms , 2012, IEEE Transactions on Visualization and Computer Graphics.

[8]  Andrew Zisserman,et al.  Very Deep Convolutional Networks for Large-Scale Image Recognition , 2014, ICLR.

[9]  Jian Sun,et al.  Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification , 2015, 2015 IEEE International Conference on Computer Vision (ICCV).

[10]  Benjamin Graham,et al.  Fractional Max-Pooling , 2014, ArXiv.

[11]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[12]  Eugenio Culurciello,et al.  ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation , 2016, ArXiv.

[13]  M. Mohammed Thaha,et al.  Brain Tumor Segmentation Using Convolutional Neural Networks in MRI Images , 2019, Journal of Medical Systems.

[14]  Pascal Fua,et al.  Exploiting Enclosing Membranes and Contextual Cues for Mitochondria Segmentation , 2014, MICCAI.

[15]  Sergey Ioffe,et al.  Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , 2015, ICML.

[16]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[17]  Trevor Darrell,et al.  Fully Convolutional Networks for Semantic Segmentation , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[18]  Luca Maria Gambardella,et al.  Deep Neural Networks Segment Neuronal Membranes in Electron Microscopy Images , 2012, NIPS.

[19]  John Salvatier,et al.  Theano: A Python framework for fast computation of mathematical expressions , 2016, ArXiv.

[20]  Nitish Srivastava,et al.  Dropout: a simple way to prevent neural networks from overfitting , 2014, J. Mach. Learn. Res..

[21]  Luc Van Gool,et al.  The Pascal Visual Object Classes (VOC) Challenge , 2010, International Journal of Computer Vision.

[22]  Roberto Cipolla,et al.  Training Convolutional Neural Networks with Low-rank Filters for Efficient Image Classification , 2016 .