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

Abstract

Cochlear implants (CIs) use electrode arrays that are surgically inserted into the cochlea to stimulate frequency-mapped nerve endings to treat patients with hearing loss. CIs are programmed postoperatively by audiologists using behavioral tests without information on electrode-cochlea spatial relationship. We have recently developed techniques to segment the intracochlear anatomy and to localize individual contacts in clinically acquired computed tomography (CT) images. Using this information, we have proposed a programming strategy that we call image-guided CI programming (IGCIP), and we have shown that it significantly improves outcomes for both adult and pediatric recipients. One obstacle to large-scale deployment of this technique is the need for manual intervention in some processing steps. One of these is the rough registration of images prior to the use of automated intensity-based algorithms. Although seemingly simple, the heterogeneity of our image set makes this task challenging. We propose a solution that relies on the automated random forest-based localization of multiple landmarks used to estimate an initial transformation with a point-based registration method. Results show that it produces results that are equivalent to a manual initialization. This work is an important step toward the full automation of IGCIP.

Keywords: image registration; landmark localization; regression forest.

Fig. 1

Computation pipelines used to segment the inner ear anatomy and to localize electrode arrays in postoperative CTs when (a) the preoperative CT scan is available and (b) when the preoperative CT scan is not available. Circles represent the atlas, preoperative (preop), or postoperative (postop) CTs. Rectangles represent operations and ellipsoids represent structures of interest. $R_{\mathrm{A}\text{-}\mathrm{pre}}$, $R_{\mathrm{A}\text{-}\text{post}}$, and $R_{\mathrm{pre}\text{-}\text{post}}$ represent the transformations from atlas to preoperative CT, atlas to postoperative CT, and preoperative CT to postoperative CT, respectively. ASM represents active shape model-based methods for segmenting inner ear anatomy given an initial segmentation projected from the atlas. The rectangles drawn with dashed lines highlight the pipeline’s final outputs.

Fig. 2

(a) Axial, (b) coronal, and (c) sagittal views of four representative image volumes in our dataset. This shows the range of coverage and orientation that can be encountered. Images of patient #3 only include the left and right temporal bones; images of patient #4 only include one temporal bone.

Fig. 3

A comparison of images acquired with a (a) Xoran xCAT^® scanner and (b) conventional scanner. Both are axial views of the ear region.

Fig. 4

An example of postoperative CT image showing the imaging artifact caused by the CI. ${\mathrm{P}}_1$, which is a voxel near the CI, has much higher intensity than ${\mathrm{P}}_2$, which is inside the bone.

Fig. 5

All seven landmarks, marked by crosses shown on (a) axial, (b) coronal, and (c) sagittal views.

Fig. 6

A flowchart of the scheme used to evaluate the proposed automatic initialization method.

Fig. 7

The red bounding box shows the ROI in an axial cross section.

Fig. 8

The axial view of the probability maps of one test CT, illustrating the multiple-maxima problem.

Fig. 9

Localization failure rate as a function of $d$.

Fig. 10

Localization failure rate as a function of $k$.