RACE-Net: A Recurrent Neural Network for Biomedical Image Segmentation

Abstract

The level set based deformable models (LDM) are commonly used for medical image segmentation. However, they rely on a handcrafted curve evolution velocity that needs to be adapted for each segmentation task. The Convolutional Neural Networks (CNN) address this issue by learning robust features in a supervised end-to-end manner. However, CNNs employ millions of network parameters which require a large amount of training data to prevent over-fitting and also increases its memory requirement and computation time during testing. Moreover, since CNNs pose segmentation as a region-based pixel labeling, they cannot explicitly model the high-level dependencies between the points on the object boundary to preserve its overall shape, smoothness or the regional homogeneity within and outside the boundary. We present a Recurrent Neural Network (RNN) based solution called the RACE-net to address the above issues. RACE-net models a generalized LDM evolving under a constant and mean curvature velocity. At each time-step, the curve evolution velocities are approximated using a feed-forward architecture inspired by the multi-scale image pyramid. RACE-net allows the curve evolution velocities to be learned in an end-to-end manner while minimizing the number of network parameters, computation time and memory requirements. The RACE-net was validated on three different segmentation tasks: Optic disc and cup in color fundus images, cell nuclei in histopathological images and the left atrium in cardiac MRI volumes. Assessment on public datasets was seen to yield high Dice values between 0.87 and 0.97 which illustrates its utility as a generic, off-the-shelf architecture for biomedical segmentation.

Joint disc and cup segmentation based on recurrent fully convolutional network

“…Deep learning methods [6,[14][15][16][17][18][19][20][21][22][23] segment OD and OC by training a large number of data samples to automatically extract features. In [15], OD and OC segmentation using superpixel classification for glaucoma screening is proposed.…”

“…These recent deep learning methods have performed well and successfully promoted the study of OD and OC segmentation of fundus images from the perspective of deep learning. In [19], the author uses RACE--Net based on a recurrent neural network to simulate a variable model of generalized level sets that evolve at constant and average curvature speeds. It can clearly simulate the high-level dependence between points on the boundary of an object, maintaining its overall shape, smoothness, or homogeneity of the area inside and outside the boundary.…”

“…At present, the segmentation methods of the OD and OC are mainly divided into the following categories: Shape-based and template matching methods [8][9][10], models based on deformable and active contours [11][12][13], and the recent deep learning methods [6,[14][15][16][17][18][19][20][21][22][23][24][25][26]. We briefly outline the existing methods below.…”

“…In order to further demonstrate the effectiveness of RFC-Net for OD and OC segmentation, on the Drishti-GS1 dataset, in this section, we compare the performance of the proposed method with BCRF [18], Superpixel [15], and Graph cut prior [12], Boosting CNN [16], U-Net [6], RACE-Net [19], Stack-U-Net [20], pOSAL [33] and several other state-of-art OD and OC segmentation methods. https://doi.org/10.1371/journal.pone.0238983.g007…”

Unsupervised Image Segmentation Algorithm using Superpixel and Cosine Similarity

“…Despite recent of supervised image segmentation is used to obtain great result of area identification, such as Q-shift Dual-Tree Complex Wavelet Transform coefficient combined with color component to do a segmentation process based on texture [1] and RACE-Net successfully used to process biomedical data segmentation [2], unsupervised image segmentation have to challenge in a promising research direction. This is related to how the human eye works to do image segmentation in an unsupervised way, on the other unsupervised image segmentation no need to have a vast variety of labeled sample dataset.…”