Segmentation of tongue muscles from super-resolution magnetic resonance images

Abstract

Imaging and quantification of tongue anatomy is helpful in surgical planning, post-operative rehabilitation of tongue cancer patients, and studying of how humans adapt and learn new strategies for breathing, swallowing and speaking to compensate for changes in function caused by disease, medical interventions or aging. In vivo acquisition of high-resolution three-dimensional (3D) magnetic resonance (MR) images with clearly visible tongue muscles is currently not feasible because of breathing and involuntary swallowing motions that occur over lengthy imaging times. However, recent advances in image reconstruction now allow the generation of super-resolution 3D MR images from sets of orthogonal images, acquired at a high in-plane resolution and combined using super-resolution techniques. This paper presents, to the best of our knowledge, the first attempt towards automatic tongue muscle segmentation from MR images. We devised a database of ten super-resolution 3D MR images, in which the genioglossus and inferior longitudinalis tongue muscles were manually segmented and annotated with landmarks. We demonstrate the feasibility of segmenting the muscles of interest automatically by applying the landmark-based game-theoretic framework (GTF), where a landmark detector based on Haar-like features and an optimal assignment-based shape representation were integrated. The obtained segmentation results were validated against an independent manual segmentation performed by a second observer, as well as against B-splines and demons atlasing approaches. The segmentation performance resulted in mean Dice coefficients of 85.3%, 81.8%, 78.8% and 75.8% for the second observer, GTF, B-splines atlasing and demons atlasing, respectively. The obtained level of segmentation accuracy indicates that computerized tongue muscle segmentation may be used in surgical planning and treatment outcome analysis of tongue cancer patients, and in studies of normal subjects and subjects with speech and swallowing problems.

Fully automated quantitative cephalometry using convolutional neural networks

“…Since the same shape model as that in the approach of Ibragimov et al [33][34][35] was used in this work, we can conclude that CNNs outperform random forests in cephalometric analysis and yield higher success detection rates for 2-, 3-, and 3.5-mm ranges. On the other hand, CNNs yield a lower success detection rate for the 4-mm range, which suggests that outliers cannot be avoided for certain cases.…”

“…The overall estimation can be further improved by refining the likelihood estimations by a probabilistic shape-based model to consider the relative spatial arrangements of the candidate estimations. For shape-based refinement, we consider the approach of modeling the spatial relationships by Gaussian kernel density estimation problems and applying random forests in a multilandmark environment that demonstrated good performance on cephalometric analysis [33][34][35] (we refer the readers to the original publications [33][34][35] for more detailed descriptions of these techniques). The overall landmark detection framework is summarized in Fig.…”

Evaluation and comparison of 3D intervertebral disc localization and segmentation methods for 3D T2 MR data: A grand challenge

“…The properties of these landmarks are studied using training images of manually segmented discs and then used for identification of the same landmarks on a new target image. The intensity appearance of each landmark is captured by Haar-like features, which proved effective for detecting landmarks from MR images of soft tissue (Ibragimov et al, 2015). To minimize landmark mis-detection, for example, when a landmark is positioned on a neighboring disc, they model the shape of the disc by measuring the pairwise spatial relationships among the landmarks.…”

The Generalized Log-Ratio Transformation: Learning Shape and Adjacency Priors for Simultaneous Thigh Muscle Segmentation

“…Gilles et al deformably register template meshes of muscles and other thigh anatomy to a novel image, matching model edges to boundaries in the images and employing regularization terms designed to model real anatomy [17], [18]. Other works have been successful in employing atlas-based segmentation to muscle segmentation (including certain thigh muscles), registering novel MRI data to an already segmented atlas [19]- [21]. Essafi et al propose a wavelet-based encoding for calf muscles represented using landmarks, which provides a hierarchical encoding of shape variability [22].…”