Statistical Shape Model for Automated Cochlear Segmentation: A Comparison of Fitting Strategies

verfasst von

Johannes Gaa, Samuel Müller, G. Jakob Lexow, Omid Majdani, Lüder Alexander Kahrs, Tobias Ortmaier

Abstract

Statistical Shape Models (SSM) became a standard tool in medical image analysis. Its versatile use led to numerous enhancements with a wide range of application possibilities. Although, the basic usage is usually the same and requires the following steps: Preparing a trainings data set, analysis of the variance of the training data, extracting a SSM, initialization of the SSM in the target image data and fitting the SSM. For the last step several strategies have been proposed. While no strategy is generally applicable, some claim to be more adaptable and others aim on application specific robustness. This work considers multiple proposed fitting methods for SSM in the context of cochlear segmentation. In the first part we give an overview of fitting algorithms and motivate a selection we examined in subsequent experiments. We used a SSM, trained by six data sets manually segmented with expert knowledge and applied it on ten new target image datasets. Each fitting per target image dataset was done using the strategies investigated. In the results section we compare those strategies regarding accuracy, robustness and runtime.

Organisationseinheit(en)

Institut für Mechatronische Systeme

Externe Organisation(en)

Medizinische Hochschule Hannover (MHH)
Exzellenzcluster Hearing4all
Vanderbilt University
Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU Erlangen-Nürnberg)

Typ

Aufsatz in Konferenzband

Seiten

137-142

Publikationsdatum

2016

Publikationsstatus

Veröffentlicht

Peer-reviewed

Ja