Deep convolutional neural networks for multi-modality isointense infant brain image segmentation

Abstract

The segmentation of infant brain tissue images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) plays an important role in studying early brain development in health and disease. In the isointense stage (approximately 6–8 months of age), WM and GM exhibit similar levels of intensity in both T1 and T2 MR images, making the tissue segmentation very challenging. Only a small number of existing methods have been designed for tissue segmentation in this isointense stage; however, they only used a single T1 or T2 images, or the combination of T1 and T2 images. In this paper, we propose to use deep convolutional neural networks (CNNs) for segmenting isointense stage brain tissues using multi-modality MR images. CNNs are a type of deep models in which trainable filters and local neighborhood pooling operations are applied alternatingly on the raw input images, resulting in a hierarchy of increasingly complex features. Specifically, we used multimodality information from T1, T2, and fractional anisotropy (FA) images as inputs and then generated the segmentation maps as outputs. The multiple intermediate layers applied convolution, pooling, normalization, and other operations to capture the highly nonlinear mappings between inputs and outputs. We compared the performance of our approach with that of the commonly used segmentation methods on a set of manually segmented isointense stage brain images. Results showed that our proposed model significantly outperformed prior methods on infant brain tissue segmentation. In addition, our results indicated that integration of multi-modality images led to significant performance improvement.

Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks

“…This includes automatic segmentation of brain lesions [2,10], tumors [9,15,21], and neuroanatomy [14,22,3], using voxelwise network architectures [14,9,17], and more recently using 3D voxelwise networks [3,10], and fully convolutional networks (FCNs) [4,13,17]. Compared to voxelwise methods, FCNs are fast in testing and training, and use the entire samples to learn local and global image features.…”

Fast and robust segmentation of the striatum using deep convolutional neural networks

“…Until now, CNN-based segmentation studies have employed patch extraction for the input of the networks (Kleesiek et al, 2016;Moeskops et al, 2016;Zhang et al, 2015). These studies were mainly aimed at skull stripping or brain tissue segmentation.…”

“…For instance, knee cartilage segmentation using deep CNN showed better performance than previous methods based on handcrafted features (Prasoon et al, 2013). Brain tissues have been accurately segmented into gray matter, white matter and cerebrospinal fluid using deep CNN (Moeskops et al, 2016;Zhang et al, 2015).…”

“…Unlike typical brain tissue segmentation, which is usually a global process for all the regions of the brain, voxels of the striatum always occupy very small proportion of the whole brain. Accordingly, the approach should be different from the previous CNN-based brain segmentation approaches (Moeskops et al, 2016;Zhang et al, 2015). Specifically, these previous approaches used a whole image as an input for CNN to segment large regions of interests.…”

Image Segmentation Using Mean Shift Based Fuzzy C-Means Clustering Algorithm: A Novel Approach

“…Because of the different segmentation methods, pixel category attribution are also different, strictly control the pixel category attribution is more difficult, so the fuzzy image segmentation method based on the obtained attention. Among them, the most representative method is the fuzzy C-means (FCM) clustering method, its advantage is that less human participation, can achieve automatic segmentation and is widely used, for example, the detection and location of military targets on the sea in order to achieve precise navigation, for the medical image to realize the positioning of tumor and lesion and cutting, recognition and detection and evaluation of the quality of agricultural products such as apple appearance [3][4][5]. However, the requirement of the segmentation precision and algorithm efficiency is requested more and more high, the traditional FCM algorithm due to the following two significant shortcomings which could not meet with the actual needs: (1) Algorithm uses only gray difference to describe node (pixel) similarity between target segmentation, the object to be susceptible to the complex structure of image features (such as texture, gray)…”