Skip to Main Content
We present a preliminary evaluation of an automated segmentation method of the quadriceps muscles from MR images of the thigh. The method is being developed to assist research into morphological properties of the quadriceps muscles as biomarkers of osteoarthritis (OA) incidence and progression. Our method uses an anatomically anchored, template-based initialization of the level set-based segmentation approach. A template image is selected using the Kullback-Leibler divergence measure based on the muscle and fat content of the thigh images. Contours of the quadriceps muscles of the chosen template are then semi-automatically registered to the image to be segmented using an affine transformation. These registered contours are used as initializations for the multi-phase level-set segmentation of the image, which is pre-processed to reduce arterial flow artifacts, the bias field, and intramuscular fat/connective tissue. Thirteen studies from eleven different subjects were analyzed. The performance was compared against manual segmentations using the Zijdenbos similarity index (ZSI). The ZSI means and standard deviations were: rectus femoris, 0.73 plusmn 0.13; vastus intermedius, 0.78 plusmn 0.09; vastus lateralis, 0.81 plusmn 0.14; vastus medialis, 0.85 plusmn 0.10.