Implementation Of Ultrasound-Based Correction System For Abdomen Soft Tissue Shift

Traditional image-guided navigation systems establish 3D model of patient's anatomical structure using 3D reconstruction and visualization of preoperative 2D images. One drawback of this approach is the modal will not reflect the change of underlying anatomical structure correctly if organs deform due to respiration and body movement, which will result in inaccurate navigation.Due to its real-time, non-invasive, and highly flexible advantages, ultrasound imaging has comprehensive clinical applications. However, its drawback is that images usually exhibit speckle and have limited region of scanning. By combining the power of both pre-operative CT and intra-operative ultrasound, it's possible to track deformed organs and observe internal structure clearly.At present, 3D ultrasound guided navigations are widely studied and applied abroad. However, they are not yet popularized in domestic clinical applications. Under such circumstance, we implement a correction system for abdomen soft tissue shift based on intra-operative 2D ultrasound. Through registration and fusion of pre-operative CT and intra-operative real-time ultrasound, we correct the deformed abdomen organs in CT. In this paper, we focus on algorithms which are adopted in several modules of correction system, as well as their implementations. These modules include ultrasound simulation, ultrasound filtering, fusion of pre-operative CT and intra-operative ultrasound.In ultrasound simulation module, we analysis and implement an algorithm which has a high degree of simulation. It takes into account both the ultrasound acquisition model and the speckle formation model. Stages including image sampling, speckle simulation, and image interpolation are introduced. When validating the effects of ultrasound filtering algorithms and multimodal registration algorithms, the use of computer synthetic images makes it possible for us to apply some objective evaluation criteria.In ultrasound filtering module, we analysis and implement mathematical morphology based methods and median filter based methods respectively. The basic requirement of ultrasound filtering is to preserve detailed information useful to later diagnosis as much as possible while suppressing speckle effectively. We obtain algorithm Open-by-reconstruction by combining basic operators like erosion and dilation in morphology, which can eliminate light points in images while preserving image's grey scale values on a general basis. In the mean time, we use decomposition of common structuring elements to accelerate filtering. Median-type filters have good ability of detail preservation, thus very appropriate to ultrasound filtering. We introduce multi-stage median filter and adaptive weighted median filter based on standard median, apply them to ultrasound images and receive good results.In CT-US fusion module, we apply iterative closest point algorithm and thin-plate splines algorithm to global rigid transformation and local non-rigid transformation respectively. Meanwhile, similarity measures including fiducial registration error (FRE), average intensity difference (SSD, SAD), normalized cross correlation (NCC) are adopted to verify the registration results. Finally, fusion of registered images is performed, and information from both modals are synthesized and displayed by adjustment of contrast and transparency.