Hippocampus Segmentation from MR Infant Brain Images via Boundary Regression
Hippocampus segmentation from MR infant brain images is indispensable for studying early brain development. However, most of hippocampus segmentation methods were developed for adult brain images, which are not suitable for infant brain images of the firs
- PDF / 943,184 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 18 Downloads / 332 Views
2
3
Nantong University, Jiangsu 226019, China Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA [email protected] Department of Computer Science, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA 4 Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, Shanghai 200240, China
Abstract. Hippocampus segmentation from MR infant brain images is indispensable for studying early brain development. However, most of hippocampus segmentation methods were developed for adult brain images, which are not suitable for infant brain images of the first year due to low image contrast and variable structural patterns of early hippocampal development. To address these challenges, we propose a boundary regression method to detect hippocampal boundaries in the infant brain images, and then use the obtained boundaries to guide the deformable segmentation. The advantages of our segmentation method are: (1) different from the recently-developed atlas-based hippocampus segmentation methods, our method does not perform time-consuming deformable registrations; (2) different from the conventional point-regression-based boundary detection methods, our boundary regression method can predict the whole hippocampal boundary by a single regression model. Experiments on MR infant brain images from 2-week-old to 1-year-old show promising hippocampus segmentation results.
1 Introduction In the first year of life, human brain undergoes a critical phase of postnatal brain development. To study brain development and detect neuro developmental disorders, identification of brain structures in MR images is a prerequisite. Among various structures, hippocampus plays an essential role in learning and memory functions of brain. Therefore, accurate hippocampus segmentation from infant brain images is highly desired for studying early brain development [1]. Currently, most of researches on hippocampus segmentation are based on atlases [2–4]. Due to the use of deformable registration between atlases and the target image, the atlas-based methods are often computationally-expensive. Besides, the existing methods are proposed mainly for hippocampus segmentation of the adult brain images. To date, it is still a challenging task to segment hippocampi in the infant brain images. The main © Springer International Publishing Switzerland 2016 B. Menze et al. (Eds.): MCV Workshop 2015, LNCS 9601, pp. 146–154, 2016. DOI: 10.1007/978-3-319-42016-5_14
Hippocampus Segmentation from MR Infant Brain Images
T1
T2
147
FA
Fig. 1. Illustration of hippocampi in T1, T2, and FA images of the same infant brain. The red and green contours indicate the left and right hippocampi, respectively. (Color figure online)
challenges include: (1) low image contrast, (2) ambiguous hippocampi boundary, and (3) small size of hippocampus with large nearby structures, as shown in Fig. 1. To address the above challenges, inspired by [5], we propose a boundary regression method to vote th
Data Loading...
