LINKS: Learning-based multi-source IntegratioN frameworK for Segmentation of infant brain images

Abstract

Segmentation of infant brain MR images is challenging due to insufficient image quality, severe partial volume effect, and ongoing maturation and myelination processes. In the first year of life, the image contrast between white and gray matters of the infant brain undergoes dramatic changes. In particular, the image contrast is inverted around 6-8 months of age, and the white and gray matter tissues are isointense in both T1- and T2-weighted MR images and thus exhibit the extremely low tissue contrast, which poses significant challenges for automated segmentation. Most previous studies used multi-atlas label fusion strategy, which has the limitation of equally treating the different available image modalities and is often computationally expensive. To cope with these limitations, in this paper, we propose a novel learning-based multi-source integration framework for segmentation of infant brain images. Specifically, we employ the random forest technique to effectively integrate features from multi-source images together for tissue segmentation. Here, the multi-source images include initially only the multi-modality (T1, T2 and FA) images and later also the iteratively estimated and refined tissue probability maps of gray matter, white matter, and cerebrospinal fluid. Experimental results on 119 infants show that the proposed method achieves better performance than other state-of-the-art automated segmentation methods. Further validation was performed on the MICCAI grand challenge and the proposed method was ranked top among all competing methods. Moreover, to alleviate the possible anatomical errors, our method can also be combined with an anatomically-constrained multi-atlas labeling approach for further improving the segmentation accuracy.

Segmentation of Craniomaxillofacial Bony Structures from MRI with a 3D Deep-Learning Based Cascade Framework

“…We also compared our method with two widely-used methods, i.e., multi-atlas based [10] (denoted as Atlas) and random forest-based [13] (denoted as LINKS, one of the widely-used learning-based methods), to further demonstrate the advantages of segmentation accuracy by our proposed method. The experimental results are shown in Fig.…”

Joint Segmentation of Multiple Thoracic Organs in CT Images with Two Collaborative Deep Architectures

“…Note that our framework (using two deep architectures) shares some similarity with a refinement step, called AutoContext Model (ACM) [13], which has been successfully applied to brain image segmentation [14], by using traditional classifiers such as Random Forests. The main idea of ACM is to iteratively refine the posterior probability maps over the labels, given not only the input features, but also the previous probabilities of a large number of context locations which provide information about neighboring organs, forcing the deep network to learn the spatial constrains for each target organ [14].…”

“…The main idea of ACM is to iteratively refine the posterior probability maps over the labels, given not only the input features, but also the previous probabilities of a large number of context locations which provide information about neighboring organs, forcing the deep network to learn the spatial constrains for each target organ [14]. In practice, this translates to train several classifiers iteratively, where each classifier is trained not only with the original image data, but also with the probability maps obtained from the previous classifier, which gives additional context information to the new classifier.…”

Fully automated esophagus segmentation with a hierarchical deep learning approach

“…We also use the probability map of the other organs (cropped in the same way) as this provides additional prior information about neighboring structures. This process is similar to the auto-context model (ACM) [15] which has been successfully applied on several segmentation tasks including brain segmentation on MRI [16]. Our proposed method is different in the sense that we use a deep learning architecture where the features are learnt by backpropagation instead of hand-crafted features as used in classical ACM.…”