Citation
Moghbel, Mehrdad
(2017)
Liver segmentation on CT images using random walkers and fuzzy c-means for treatment planning and monitoring of tumors in liver cancer patients.
Doctoral thesis, Universiti Putra Malaysia.
Abstract
Liver and liver tumor segmentation from Computed Tomography (CT) images and
tumor burden analysis play an important role in the choice of therapeutic strategies for
liver tumor treatment planning and monitoring. Furthermore, accurate segmentation of
the blood vasculature structure can result in improved diagnosis and easier liver tumor
treatment planning. Although many methods have been proposed, this segmentation
remains a challenging task due to the lack of visible edges on most boundaries of the
liver coupled with high variability of both intensity patterns and anatomical appearances
in pathological livers. In this thesis, a new segmentation method for liver, liver tumors
and liver vasculature structure from contrast-enhanced CT imaging is proposed. As
manual segmentation for liver treatment planning is both labor intensive and timeconsuming,
a more accurate and automatic segmentation is desired. The proposed
method is fully automatic, requiring no user interaction. The proposed segmentation is
evaluated on real-world clinical data using publicly accessible benchmark clinical liver
datasets containing one of the highest numbers of tumors and pathological livers utilized
for liver tumor and vasculature segmentation. The proposed method is based on a hybrid
method integrating random walkers algorithm with integrated priors and particle swarm
optimized spatial fuzzy c-means (FCM) algorithm with level set method and AdaBoost
classifier.
Based on the location of the lung, the liver dome is automatically detected and the liver
is then extracted by random walkers method and refined using a fuzzy level set method.
This is followed by the clustering of the liver tissues using particle swarm optimized
spatial FCM algorithm. Then, these tissues are classified into tumors and blood vessels
by an AdaBoost classification method based on tissue features extracted utilizing first,
second and higher order image features selected by a minimal-redundancy maximalrelevance
feature selection approach. Finally, the segmentation is refined using level set
method.The proposed method is able to segment all tumors and blood vessels with a largest
axial diameter of over 5mm and 3mm, respectively. In comparison, RECIST standard
commonly used in evaluation of tumors suggests a minimum largest axial diameter of
10mm for tumors and has no recommendations for the minimum largest axial diameter
of blood vessels. The proposed method showed high accuracy on segmenting livers with
a mean overlap error of 6.3% and mean absolute relative volume difference of 1.9%. In
the case of liver tumor segmentation, with a mean overlap error of 19.6% and mean
absolute relative volume difference of 11.2%. For liver vasculature segmentation, a
mean overlap error of 35.9% and mean absolute relative volume difference of 16.7%
was achieved by using the proposed segmentation method, making it amongst the most
accurate liver vasculature segmentation methods. The medical applicability of the
proposed method was further validated by a consultant radiologist where in 80% of the
segmented tumors, the segmentation by the proposed method was preferred over the
provided ground truth segmentation in the dataset. In case of liver envelope
segmentation, the consultant radiologist suggested that the liver segmentation by the
proposed method is clinically acceptable.
Download File
Additional Metadata
Actions (login required)
![View Item View Item](/style/images/action_view.png) |
View Item |