Multi-Scale deep learning framework for cochlea localization, segmentation and analysis on clinical ultra-high-resolution CT images.

Minimally invasive surgery is often built upon a time-consuming preoperative step consisting of segmentation and trajectory planning. At the temporal bone, a complete automation of these two tasks might lead to faster interventions and more reproducible results, benefiting clinical workflow and patient health. We propose an automatic segmentation and trajectory planning pipeline for image-guided interventions at the temporal bone. For segmentation, we use a shape regularized deep learning approach that is capable of automatically detecting even the cluttered tiny structures specific for this anatomy. We then perform trajectory planning for both linear and nonlinear interventions on these automatically segmented risk structures. We evaluate the usability of segmentation algorithms for planning access canals to the cochlea and the internal auditory canal on 24 CT data sets of real patients. Our new approach achieves similar results to the existing semiautomatic method in terms of Dice but provides more accurate organ shapes for the subsequent trajectory planning step. The source code of the algorithms is publicly available. Automatic segmentation and trajectory planning for various clinical procedures at the temporal bone are feasible. The proposed automatic pipeline leads to an efficient and unbiased workflow for preoperative planning.

Toward an automatic preoperative pipeline for image-guided temporal bone surgery

Temporal bone CT-scan is a prerequisite in most surgical procedures concerning the ear such as cochlear implants. The 3D vision of inner ear structures is crucial for diagnostic and surgical preplanning purposes. Since clinical CT-scans are acquired at relatively low resolutions, improved performance can be achieved by registering patient-specific CT images to a high-resolution inner ear model built from accurate 3D segmentations based on micro-CT of human temporal bone specimens. This paper presents a framework based on convolutional neural network for human inner ear segmentation from micro-CT images which can be used to build such a model from an extensive database. The proposed approach employs an auto-context based cascaded 2D U-net architecture with 3D connected component refinement to segment the cochlear scalae, semicircular canals, and the vestibule. The system was formulated on a data set composed of 17 micro-CT from public Hear-EU dataset. A Dice coefficient of 0.90 and Hausdorff distance of 0.74 mm were obtained. The system yielded precise and fast automatic inner-ear segmentations.

/pdf/automatic-segmentation-of-inner-ear-on-ct-scan-using-auto-58mdkedwd5.pdf

Automatic segmentation of inner ear on CT-scan using auto-context convolutional neural network.

Cochlear implants (CI) are used to treat hearing loss by surgically inserting an electrode array into the organ of hearing, the cochlea. Pre- and post-operative CT images are used routinely for surgery planning and evaluation of cochlear implantation. However, due to the small size and the complex topology of the cochlea, the anatomical information that can be extracted from the images is limited. The first focus of this work aims at defining automatic image processing methods adapted to the spiral shape of the cochlea to study the cochlear shape variability from high-resolution μCT images. The second focus aims at developing and evaluating a new parametric cochlear shape model. The model is applied to extract patient-specific clinically relevant metrics such as the maximal insertion depth of CI electrode arrays. Thanks to the uncertainty quantification, provided by the model, we can assess the reliability of CT-based segmentation as compared to the ground truth segmentation provided by μCT scans. Finally, the last focus concerns a joint model of the cochlear shape (and its substructures) model and its appearance within a generative probabilistic Bayesian framework. The proposed segmentation method was applied to a large database of 987 CT images and allowed the statistical characterization of the cochlear anatomical variability along with the quantification of the bilateral symmetry. This work paves the way to novel clinical applications such as improved diagnosis by identifying pathological cochlear shapes; preoperative optimal electrode design and insertion axis planning; postoperative electrode position estimation and implantation evaluation; and cochlear implantation simulation.

https://tel.archives-ouvertes.fr/tel-01609910/document

Segmentation and study of anatomical variability of the cochlea from medical images

A cochlear implant is an electronic device which can restore sound to completely or partially deaf patients. For surgical planning, a patient-specific model of the inner ear must be built using high-resolution images accurately segmented. We propose a new framework for segmentation of micro-CT cochlear images using random walks, where a region term estimated by a Gaussian mixture model is combined with a shape prior initially obtained by a statistical shape model (SSM). The region term can then take advantage of the high contrast between the background and foreground, while the shape prior guides the segmentation to the exterior of the cochlea and to less contrasted regions inside the cochlea. The prior is obtained via a non-rigid registration regularized by a statistical shape model. The SSM constrains the inner parts of the cochlea and ensures valid output shapes of the inner ear.

/pdf/random-walks-with-statistical-shape-prior-for-cochlea-and-2vd6otc5jx.pdf

Random walks with statistical shape prior for cochlea and inner ear segmentation in micro-CT images

Cochlear implants can restore hearing to deaf or partially deaf patients. In order to plan the intervention, a model from high resolution µCT images is to be built from accurate cochlea segmentations and then, adapted to a patient-specific model. Thus, a precise segmentation is required to build such a model. We propose a new framework for segmentation of µCT cochlear images using random walks where a region term is combined with a distance shape prior weighted by a confidence map to adjust its influence according to the strength of the image contour. Then, the region term can take advantage of the high contrast between the background and foreground and the distance prior guides the segmentation to the exterior of the cochlea as well as to less contrasted regions inside the cochlea. Finally, a refinement is performed preserving the topology using a topological method and an error control map to prevent boundary leakage. We tested the proposed approach with 10 datasets and compared it with the latest techniques with random walks and priors. The experiments suggest that this method gives promising results for cochlea segmentation.

/pdf/cochlea-segmentation-using-iterated-random-walks-with-shape-4j650tdkip.pdf

Cochlea segmentation using iterated random walks with shape prior

Cochlear implantation is a safe and effective surgical procedure to restore hearing in deaf patients. However, the level of restoration achieved may vary due to differences in anatomy, implant type and surgical access. In order to reduce the variability of the surgical outcomes, we previously proposed the use of a high-resolution model built from $$\mu \hbox {CT}$$
 images and then adapted to patient-specific clinical CT scans. As the accuracy of the model is dependent on the precision of the original segmentation, it is extremely important to have accurate $$\mu \hbox {CT}$$
 segmentation algorithms. We propose a new framework for cochlea segmentation in ex vivo $$\mu \hbox {CT}$$
 images using random walks where a distance-based shape prior is combined with a region term estimated by a Gaussian mixture model. The prior is also weighted by a confidence map to adjust its influence according to the strength of the image contour. Random walks is performed iteratively, and the prior mask is aligned in every iteration. We tested the proposed approach in ten $$\mu \hbox {CT}$$
 data sets and compared it with other random walks-based segmentation techniques such as guided random walks (Eslami et al. in Med Image Anal 17(2):236–253, 2013) and constrained random walks (Li et al. in Advances in image and video technology. Springer, Berlin, pp 215–226, 2012). Our approach demonstrated higher accuracy results due to the probability density model constituted by the region term and shape prior information weighed by a confidence map. The weighted combination of the distance-based shape prior with a region term into random walks provides accurate segmentations of the cochlea. The experiments suggest that the proposed approach is robust for cochlea segmentation.

Random walks with shape prior for cochlea segmentation in ex vivo \mu \hbox {CT}

Iterated random walks with shape prior

Cochlear implants can restore hearing to completely or partially deaf patients. The intervention planning can be aided by providing a patient-specific model of the inner ear. Such a model has to be built from high resolution images with accurate segmentations. Thus, a precise segmentation is required. We propose a new framework for segmentation of micro-CT cochlear images using random walks combined with a statistical shape model (SSM). The SSM allows us to constrain the less contrasted areas and ensures valid inner ear shape outputs. Additionally, a topology preservation method is proposed to avoid the leakage in the regions with no contrast.

Esmeralda Ruiz Pujadas

Papers

Cochlea segmentation using iterated random walks with shape prior

Random walks with shape prior for cochlea segmentation in ex vivo \mu \hbox {CT}

Random walks with statistical shape prior for cochlea and inner ear segmentation in micro-CT images

Iterated random walks with shape prior

Statistical Shape Model with Random Walks for Inner Ear Segmentation