University of Cambridge Home Department of Engineering

F-AHG-3: Segmenting the cochlea in clinical CT

This project is motivated by cochlear implants. The precise positioning of the implant within the cochlea can have a profound affect on the hearing outcome. And yet it is difficult for the surgeon to take the individual's particular cochlear size and shape into account, since clinical CT imaging is unable to resolve this tiny structure in detail.

A companion project will look to build a statistical shape model of the human cochlea. The source data will be high resolution, micro-CT scans of cadaveric temporal bones. These scans reveal the detailed cochlear geometry, though segmenting the cochlear duct and semicircular canals is a challenge. Given a sufficient number of cochleas, properly segmented, principal component analysis can be used to reveal the significant modes of shape variation in the population.

The idea is to then fit the shape model to low resolution, clinical scans of cochleas. Even though there is little detail in clinical scans, the fitted model should reveal the size and gross shape of the individual cochlea, which will provide valuable information to the surgeon performing the implant.

This is one of two linked projects, offered in collaboration with Professor Manohar Bance at Addenbrooke's Hospital. Project F-AHG-3 will focus on fitting the shape model to the clinical CT data. The project would suit a student who has taken Module 3G4 and Project GG2, though neither is a strict prerequisite. It will involve programming in C++, and the opportunity to learn more about computational geometry and graphics. See the work of Noble et al for further details of what we are trying to achieve, though our methods are likely to be different.

Micro-CT scans of the human cochlea offer rich detail (left). By segmenting the key anatomical structures (right), the size and shape of the cochlea is revealed: this particular specimen has been fitted with a cochlear implant. Manual segmentation is laborious and time-consuming. Project F-AHG-2 will investigate techniques for speeding up the segmentation process.
Given sufficient segmented cochleas, principal component analysis can be used to build a statistical shape model, revealing the significant modes of variation in the population. The example here is the third shape mode in a pilot study of eight cochleas. The "roller-coaster" (down then up) height profile of some cochleas potentially affects implant insertion.
Clinical CT scans of human heads are much lower resolution and reveal little cochlear detail. But it should be possible to fit the statistical shape model to this data. The way the model fits will inform the surgeon of the cochlea's gross size and shape, for example whether it has the "roller-coaster" profile illustrated above. Project F-AHG-3 will investigate techniques for fitting the model to the clinical CT data.
© Cambridge University Engineering Department
Information provided by Andrew Gee
Last updated: March 2018