Patient-specific estimation of detailed cochlear shape from clinical CT images

Abstract

The paper presents the process of building and using the SDM of the cochlea. Compared to current best practice, we demonstrate competitive performance and some useful properties of our method.

Practicable assessment of cochlear size and shape from clinical CT images

“…There are also likely differences in the way the gold-standard and clinical coordinate systems were aligned. Kjer et al 23 reported mean surface errors of 0.11 mm for the cochlear scalae.…”

“…Evaluation metrics were based on Dice similarity coefficients and surface errors, so it is unclear how well this method can estimate surgically relevant parameters. Kjer et al 23 developed a similar approach using a statistical deformation model, reporting measurement accuracy and precision for cochlear length, width and height in addition to surface errors, but with no consideration of vertical trajectories. van der Jagt et al 24 describe an automatic, three-dimensional tracing method that was used to estimate inner and outer wall radii, duct diameter and vertical trajectory in low-resolution CT scans of 242 patients.…”

“…In this work, we describe a fast, simple and freely available method to fit surface models of the otic capsule to CT data. The fitting may be directed by a statistical model, in the spirit of Noble et al 22 and Kjer et al 23 , or constrained only by a smoothness criterion, in the spirit of Iyaniwura et al 10 . We compare the performance of the two approaches, with specific reference to three-dimensional, surgically relevant measurements like those considered by van der Jagt et al 24 .…”

Morphological analysis of sigmoid sinus anatomy: Clinical applications to neurotological surgery

“…These algorithms can then be used to build anatomical models from clinical CT imaging, allowing accurate 3D reconstruction of a patient’s anatomy. Creation of automatic segmentation algorithms requires tedious manual segmentation of micro-CT data, therefore many groups have been working on automating segmentation with polynomial functions [10], atlas-based registration [11], SSMs [12–14], and deep learning [15, 16]. Further, labour intensive manual segmentation often leads to smaller sample sizes.…”

“…Segmentation of these structures in 3D is important for surgical planning [3], robotic surgery [4], virtual-reality surgical simulation [5–7], and patient-specific cochlear implant programming [8, 9]. Unfortunately, manual segmentation is very labour intensive, therefore many groups have been working on automating segmentation with polynomial functions [10], atlas-based registration [11], statistical shape models [12–14], and deep learning [15, 16]. This type of research requires large datasets, and many groups have made their datasets publicly available to help the larger research community [17].…”