Localizing landmark sets in head CTs using random forests and a heuristic search algorithm for registration initialization

HeadLocNet: Deep convolutional neural networks for accurate classification and multi-landmark localization of head CTs

Cochlear implants (CIs) are used to treat subjects with hearing loss. In a CI surgery, an electrode array is inserted into the cochlea to stimulate auditory nerves. After surgery, CIs need to be programmed. Studies have shown that the cochlea-electrode spatial relationship derived from medical images can guide CI programming and lead to significant improvement in hearing outcomes. We have developed a series of algorithms to segment the inner ear anatomy and localize the electrodes. But, because clinical head CT images are acquired with different protocols, the field of view and orientation of the image volumes vary greatly. As a consequence, visual inspection and manual image registration to an atlas image are needed to document their content and to initialize intensity-based registration algorithms used in our processing pipeline. For large-scale evaluation and deployment of our methods these steps need to be automated. In this article we propose to achieve this with a deep convolutional neural network (CNN) that can be trained end-to-end to classify a head CT image in terms of its content and to localize landmarks. The detected landmarks can then be used to estimate a point-based registration with the atlas image in which the same landmark set's positions are known. We achieve 99.5% classification accuracy and an average localization error of 3.45 mm for 7 landmarks located around each inner ear. This is better than what was achieved with earlier methods we have proposed for the same tasks.

Automatic graph-based method for localization of cochlear implant electrode arrays in clinical CT with sub-voxel accuracy

Cochlear implants (CIs) are neural prosthetics that provide a sense of sound to people who experience severe to profound hearing loss. Recent studies have demonstrated a correlation between hearing outcomes and intra-cochlear locations of CI electrodes. Our group has been conducting investigations on this correlation and has been developing an image-guided cochlear implant programming (IGCIP) system to program CI devices to improve hearing outcomes. One crucial step that has not been automated in IGCIP is the localization of CI electrodes in clinical CTs. Existing methods for CI electrode localization do not generalize well on large-scale datasets of clinical CTs implanted with different brands of CI arrays. In this paper, we propose a novel method for localizing different brands of CI electrodes in clinical CTs. We firstly generate the candidate electrode positions at sub-voxel resolution in a whole head CT by thresholding an up-sampled feature image and voxel-thinning the result. Then, we use a graph-based path-finding algorithm to find a fixed-length path that consists of a subset of the candidates as the localization result. Validation on a large-scale dataset of clinical CTs shows that our proposed method outperforms the state-of-art CI electrode localization methods and achieves a mean error of 0.12 mm when compared to expert manual localization results. This represents a crucial step in translating IGCIP from the laboratory to large-scale clinical use.

Automatic localization of closely spaced cochlear implant electrode arrays in clinical CTs

PURPOSE

Cochlear implants (CIs) are neural prosthetic devices that provide a sense of sound to people who experience profound hearing loss. Recent research has indicated that there is a significant correlation between hearing outcomes and the intracochlear locations of the electrodes. We have developed an image-guided cochlear implant programming (IGCIP) system based on this correlation to assist audiologists with programming CI devices. One crucial step in our IGCIP system is the localization of CI electrodes in postimplantation CTs. Existing methods for this step are either not fully automated or not robust. When the CI electrodes are closely spaced, it is more difficult to identify individual electrodes because there is no intensity contrast between them in a clinical CT. The goal of this work is to automatically segment the closely spaced CI electrode arrays in postimplantation clinical CTs.

METHODS

The proposed method involves firstly identifying a bounding box that contains the cochlea by using a reference CT. Then, the intensity image and the vesselness response of the VOI are used to segment the regions of interest (ROIs) that may contain the electrode arrays. For each ROI, we apply a voxel thinning method to generate the medial axis line. We exhaustively search through all the possible connections of medial axis lines. For each possible connection, we define CI array centerline candidates by selecting two points on the connected medial axis lines as the array endpoints. For each CI array centerline candidate, we use a cost function to evaluate its quality, and the one with the lowest cost is selected as the array centerline. Then, we fit an a priori known geometric model of the array to the centerline to localize the individual electrodes. The method was trained on 28 clinical CTs of CI recipients implanted with three models of closely spaced CI arrays. The localization results are compared with the ground truth localization results manually generated by an expert.

RESULTS

A validation study was conducted on 129 clinical CTs of CI recipients implanted with three models of closely spaced arrays. Ninety-eight percent of the localization results generated by the proposed method had maximum localization errors lower than one voxel diagonal of the CTs. The mean localization error was 0.13 mm, which was close to the rater's consistency error (0.11 mm). The method also outperformed the existing automatic electrode localization methods in our validation study.

CONCLUSION

Our validation study shows that our method can localize closely spaced CI arrays with an accuracy close to what is achievable by an expert on clinical CTs. This represents a crucial step toward automating IGCIP and translating it from the laboratory to the clinical workflow.

Accurate Detection of Inner Ears in Head CTs Using a Deep Volume-to-Volume Regression Network with False Positive Suppression and a Shape-Based Constraint

Cochlear implants (CIs) are neural prosthetics which are used to treat patients with hearing loss. CIs use an array of electrodes which are surgically inserted into the cochlea to stimulate the auditory nerve endings. After surgery, CIs need to be programmed. Studies have shown that the spatial relationship between the intra-cochlear anatomy and electrodes derived from medical images can guide CI programming and lead to significant improvement in hearing outcomes. However, clinical head CT images are usually obtained from scanners of different brands with different protocols. The field of view thus varies greatly and visual inspection is needed to document their content prior to applying algorithms for electrode localization and intra-cochlear anatomy segmentation. In this work, to determine the presence/absence of inner ears and to accurately localize them in head CTs, we use a volume-to-volume convolutional neural network which can be trained end-to-end to map a raw CT volume to probability maps which indicate inner ear positions. We incorporate a false positive suppression strategy in training and apply a shape-based constraint. We achieve a labeling accuracy of 98.59% and a localization error of 2.45 mm. The localization error is significantly smaller than a random forest-based approach that has been proposed recently to perform the same task.