Paper Review:-LADDERNET:MULTI-PATH NETWORKS BASED ON U-NET FOR MEDICAL IMAGE SEGMENTATION
Let us look at its key points and why it makes more sense than existing UNets from this paper
U-Net has been providing state-of-the-art performance in many medical image segmentation problems. Many modifications have been proposed for U-Net, such as attention U-Net, recurrent residual convolutional U-Net (R2-UNet), and U-Net with residual blocks or blocks with dense connections. However, all these modifications have an encoderdecoder structure with skip connections, and the number of paths for information flow is limited.
Laddernet can be viewed as a chain of multiple U-nets.
Features in different spatial scales are named with letters A to E, and columns are named with numbers 1 to 4. 1 and 3 are named as encoder branches while 2 and 4 are named as decoder branches.
As we go from smaller receptive field features to larger (eg. A to B),we use a conv layer with a stride 2 and likewise for going from larger to smaller receptive field features, we use a conv_transpose with stride 2.
What’s the benefit over Unet?
LadderNet provides multiple paths of information flow, and we lists a few paths here as an example: (1) A1 → A2 → A3 → A4, (2) A1 → A2 → A3 → B3 → B4 → A4, (3) A1 → B1 → B2 → B3 → B4 → A4, (4) A1 → B1 → C1 → D1 → E1 → D2 → C2 → B2 → A2 → A3 → A4.
Each path can be viewed as a variant of FCN. LadderNet can also be viewed as an ensemble of multiple FCNs.LadderNet has the potential to capture more complicated features and produce a higher accuracy.So the architecture proposed here in this paper is “just enough” diverse to allow capturing complicated features while reducing the total number of network parameters(1.5Million here). Probably GPipe can help us optimize this more.
Why the ‘shared weights’ residual block is used here?
More encoder-decoder branches will increase the number of parameters and the difficulty of training. To solve this problem, we propose shared-weights residual blocks as shown.
Different from a standard residual convolutional block, the two convolutional layers in the same block share the same weights. Similar to the recurrent convolutional neural network (RCNN), the two convolutional layers in the same block can be viewed as one recurrent layer, except that the two batch normalization layers are different. A drop-out layer is added between two convolutional layers to avoid overfitting. The shared-weights residual block combines the strength of skip connection, recurrent convolution and drop-out regularization, and has much fewer parameters that a standard residual block.
Dataset used and Evaluation Metrics
AUC-ROC score has been used for evaluation.
Hyperparameters and Parameters
- Adam Optimizer
- ReduceLROnPlateau
- Lr Scheduler with lrs=1e-2,1e-3,1e-4 at epochs 0 ,20, 150 respectively
- Num of epochs = 250
- BatchSize = 1024
- Loss fn =vanilla Crossentropy
- dropout = 0.25
Thank you for reading!
Sayantan Das,
https://www.linkedin.com/in/sayantan-das-95b50a125/
References
[1] Yann LeCun et al., “Deep learning,” nature, 2015.
[2] Alex Krizhevsky et al., “Imagenet classification with deep convolutional neural networks,” in Advances in neural information processing systems, 2012.
[3] Jia Deng et al., “Imagenet: A large-scale hierarchical image database,” in CVPR 2009., 2009.
[4] Jonathan Long et al., “Fully convolutional networks for semantic segmentation,” in CVPR, 2015.
[5] Olaf Ronneberger et al., “U-net: Convolutional networks for biomedical image segmentation,” in MICCAI, 2015.
[6] Hengshuang Zhao et al., “Pyramid scene parsing network,” in (CVPR), 2017.
[7] Liang-Chieh Chen et al., “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,” IEEE transactions on pattern analysis and machine intelligence, 2018.
[8] Md Zahangir Alom et al., “Recurrent residual convolutional neural network based on u-net (r2u-net) for medical image segmentation,” 2018.
[9] Simon Jegou et al., “The one hundred layers tiramisu: ´ Fully convolutional densenets for semantic segmentation,” in (CVPRW), 2017.
[10] Andreas Veit et al., “Residual networks behave like ensembles of relatively shallow networks,” in Advances in Neural Information Processing Systems, 2016.
[11] Kaiming He et al., “Deep residual learning for image recognition,” in CVPR, 2016.
[12] Md Zahangir Alom et al., “Inception recurrent convolutional neural network for object recognition,” 2017.
[13] Joao VB Soares et al., “Retinal vessel segmentation using the ˜ 2-d gabor wavelet and supervised classification,” IEEE Transactions on medical Imaging, 2006.
[14] Sohini Roychowdhury et al., “Blood vessel segmentation of fundus images by major vessel extraction and subimage classification,” JBHI, 2015.
[15] Qiaoliang Li et al., “A cross-modality learning approach for vessel segmentation in retinal images.,” IEEE Trans. Med. Imaging, 2016.
