Coupled Deformable Model For Bladder Wall Segmentation

The accurate segmentation of bladder wall in magnetic resonance (MR) imageplays a key role in both clinical application and medical research. Bladder cancer is aserious disease with high incidence and recurrence rate. Thus earlier detection iscrucial. The abnormal thickening of the bladder wall can be used as an importantindicator of the bladder tumor. The most fundamental and important work to measurethe thickness of bladder wall is to segment the inner and outer bladder boundariesaccurately. At present, in the clinical application, this work is usually done manuallyby the medical workers. However, it is a heavy and time consuming work. Especiallywhen the processing images increase greatly under the development of medicalimaging technology, it is unable to be accomplished by only relying on manualdelineation. Therefore, in this paper, we aim to propose an automatic accurate methodfor bladder wall segmentation.To address the challenges of the bladder wall segmentation in MR image, weresearched and proposed several effective methods. First, for the artifacts in thebladder MR image, the gradient direction is exploited to propose a directional levelset model to distinguish the inner boundary from the artifacts edges, which reducesthe influences of the artifacts at some extent. For the complicated outside tissues, weexploit the region information of the bladder wall to construct a coupled level setmodel, which can segment the inner and outer boundaries simultaneously and refinethe outer segmentation by the inner segmentation. Moreover, a prior knowledge of theminimum thickness of bladder wall is added to the coupled model to avoid the overlapor cross between inner and outer zero level sets. Second, for the partial weakboundaries in the slices closed to the base and the apex areas of bladder, we use thesegmentation of the previous slice as the shape prior to adaptively constrain thesegmentation of the current slice, which initially solves the leakage of level set in theweak boundary. As the shape prior has been proven to be effective in the bladder wallsegmentation, we further propose a more accurate shape prior construction method, partial sparse shape model, which uses the partial reliable contour to construct acomplete reliable shape prior. A sector-driven level set model is also proposed to meetthe different needs of shape constraint in different region and different evolution stage.Finally，the proposed partial sparse shape model is extended to the classic activeshape model (ASM) method to address the incorrect searched results caused by thepartial weak boundary, which demonstrates the universality and effectiveness of theproposed model.The main contribution of this paper can be summarized as follows:1) Acoupled directional level set model;2) An adaptive shape constrained level set model;3) A partial sparse shape constrained sector-driven level set model;4) The extensionof the proposed partial sparse shape model on the classical ASM method. Ourmethods are tested on15different patients with overall167slices, the obtained results:the inner P2C is1.06±0.28mm, inner DSC is0.98±0.01, and the outer P2C is1.46±0.42mm, outer DSC is0.97±0.01. Compared to the state-of-the-art methods,our methods are demonstrated to be effective and accurate.