Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article


Master of Engineering Science


Electrical and Computer Engineering


Ladak, Hanif M.

2nd Supervisor

Agrawal, Sumit K.



Medical image segmentation is an important step to identify the shape and position of patient anatomy prior to surgical simulation, surgical rehearsal, and surgical planning. It is crucial that the facial nerve (FN) is segmented accurately as damage to this nerve can severely impact facial expression, speech, and taste. Manual segmentation provides accurate results but is time-consuming and labor-intensive; semi-automatic methods of segmentation are more feasible in a clinical setting and can provide accurate results with minimal user involvement. The objective of this work was to create a novel, open-source, multi-atlas based segmentation algorithm of the entire FN requiring minimal user intervention. Twenty-eight temporal bones were segmented producing an average Dice metric of 0.76 and an average Hausdorff distance of 0.17 mm which is similar to previously published algorithms. These results indicate that this segmentation approach can accurately segment the FN and greatly reduce time spent with manual segmentation.

Summary for Lay Audience

Surgeries can be difficult since important structures must be avoided and every person has different anatomy. Due to the differences between people, a process called medical image segmentation is needed. Segmentation is used to find the shape and position of a person’s anatomy and label it. These labels can then be used for simulation, rehearsal, and surgical planning. One structure which should be labelled is the facial nerve, which is found in the temporal bone. The facial nerve is hard to identify on medical images since it is small, and it can be hard to tell apart from nearby anatomy due a lack of contrast. Another reason it should be labelled is that the shape and position of the facial nerve changes between any two people. If the facial nerve is not labelled correctly, the facial nerve can be damaged. This can cause issues with facial expression, speech, and taste, which can have a terrible impact on a patient's life.
Manually labeling the facial nerve can provide good results but takes more time and effort, making it a poor choice for a clinical setting. A preferred way to label the facial nerve is through semi-automatic methods. These methods are useful in a clinical setting because they can provide good results with less effort. Algorithms already exist to label the facial nerve, but none of them label the whole facial nerve, and most are not openly available to use. The aim of this work was to create an openly available algorithm to label the whole facial nerve. This is done using atlas-based techniques. These techniques make use of facial nerves which have already been labelled, to fully label new facial nerves, with very little effort. This algorithm was tested by labeling the facial nerve in 28 temporal bones. The results showed that the labels this algorithm produced were very close to the actual facial nerves. The results were also very close to the results shown in previous work, while using less time than a manual segmentation would.

Included in

Biomedical Commons