Active Contour Kidney Segmentation Based On A 3D DCE-MRI Time Series

Nowadays, an increasing number of people are suffering from renaldiseases, and clinical doctors call for several functional parameters toevaluate them. As one important parameter, the renal volume is vital inmany clinical setting, such as potential kidney donor evaluation andmaybe an indicator of potential functional residual capacity. In addition totime and labor consuming manual delineation methods, morecomputer-aided non-model and model-based algorithms are being appliedto image segmentation.In this thesis, we aim at achieving a fully automatic segmentationapproach based on a 3D DCE-MRI perfusion time series to acquire renalarea of each slice and volume assessment for each dataset.One single ordinary MRI kidney image has low contrast among thecompartments such as cortex, medulla and pelvis because of the similardensity of the hydrogen in the body fluids. After some certainparamagnetic contrast agent was injected into the kidney, it flew into thecortex, medulla and pelvis, changing the intensity over time andenhancing the contrast.According to this property of intensity variance in differentcompartments, k-means clustering was first used to classify the kidneyinto three regions and initialize the kidney contour, followed by a discreteparametric active contour (snake) to obtain the outer boundary of thekidney. In this modified snake model, the traditional continuouscomponent was removed and two additional distance images were addedto guarantee a robust lock of the snake to the actual kidney contour.Our method was tested on eight 3D DCE-MRI renal datasets (four leftkidneys, four right kidneys) of healthy volunteers. The segmentationresults were evaluated by comparison with manual delineated volumes.The deviation results of the delineated contours by our approach of theleft kidney are 6.14%, 5.05%, 8.04% and 6.40%, while the results of the right kidney are 4.44%, 6.11%, 6.78% and 4.16%. The average is 5.88%.