Skip to main content

The Use of Fast Marching Methods in Medical Image Segmentation

Varduhi Yeghiazaryan and Irina Voiculescu


Despite several decades of research into segmentation techniques, unsupervised medical image segmentation is barely usable in a clinical context, and still at vast user time expense. The Fast Marching method is an established segmentation technique for generic spaces, which typically requires manual initialisation. This paper illustrates unsupervised organ segmentation through the use of a novel automated labelling algorithm followed by a hypersurface front propagation method — Fast Marching. The labelling stage relies on a pre-computed image partition forest obtained directly from CT scan data. We perform a systematic analysis of the effects of the Fast Marching method parameters, and compare the performance of the algorithm in different settings for a specific task. We also introduce novel approaches to the choice of some parameters of the Fast Marching relying on the results of hierarchical image segmentation algorithms. We have implemented all procedures to operate directly on 3D volumes, rather than slice–by–slice, because our algorithms are dimensionality– independent. The results picture segmentations which identify abdominal organs (such as the liver and kidneys), but can easily be extrapolated to other body parts. Quantitative analysis of our unsupervised segmentation compared against hand–segmented gold standards for kidney segmentation indicates an average Dice similarity coefficient of 90%. Results were obtained over volumes of CT data with 9 kidneys, computing both volume–based similarity measures (such as the Dice and Jaccard coefficients, true positive volume fraction) and size–based measures (suchas the relative volume difference). Our analysis considers both healthy and diseased kidneys, although extreme pathological cases were excluded from the overall count. Such cases are difficult to segment both manually and automatically due to the large amplitude of Hounsfield unit distribution in the scan, and the wide spread of the tumorous tissue inside the abdomen. In the case of kidneys that have maintained their shape, the similarity range lies around the values obtained for inter–operator variability. Whilst the procedure is fully unsupervised, our tools also provide a light level of manual editing.

Oxford‚ UK
Department of Computer Science