Lymph node segmentation in CT images using a size invariant Mass Spring Model

One major challenge in automated segmentation of lymph nodes in CT scans is the high variance in terms of texture, surrounding tissue, shape and also size. Mass Spring Models have been proven to be suitable for this task. However due to their size preserving property, their performance is highly affected by the size of the target structures. This paper addresses this point by introducing a size invariant Mass Spring Model, which relates to relative rest lengths, has balanced torsion forces and an initial model expansion. We evaluated our method on a set of 25 lymph nodes from routinely gathered CT images and compared it to state of the art Mass Spring Models with different initial sizes. The average Dice Similarity Coefficient toward gold standard was 0.72 for our method compared to 0.61 for the best fitted state of the art model. Thus our method can be successfully applied to clinical relevant lymph nodes of different size without prior knowledge about the size of the target structures in contrast to existing methods.