Application Research Of Image Segmentation & Image Registration For Brachytherapy

Radiation therapy is one of the key techniques for cancer treatment, and above 70% of cancer patients need radiation therapy. The radiation therapy is aim to maximize the gain ratio of radiation therapy, which means focusing the maximized dose on the planning target volume (PTV) and protecting the normal tissue and important organs (OARs) from unnecessary irradiation.Image segmentation and image registration are the key techniques in the field of radiation therapy at present. Firstly, the segmentation or contouring of PTV and OAR affect the accuracy of radiation therapy directly. Second, we need to deform the planning CT image to the CT or CBCT image before therapy in the clinic application of image guided radiation therapy (IGRT), which could be used to obtain the position change and correct the patient’s position. In adaptive radiation therapy, extrapolating the contour of PTV and OARs,the dose accumulation in multi-fractional radiation therapy and the replanning of radiation therapy planning are all rely on the Deformable Image Registration (DIR). The accuracy of the image registration could influence the radiation therapy effect directly.Not only for 3-dimensional conformal radiation therapy (3D-CRT), but also for intensity modulated radiation therapy (IMRT). In the therapy planning process, the present radiation therapy technology needs the physician contouring the tumor region and the OAR before treatment, and making the treatment planning through the treatment planning system (TPS). The accuracy of contouring the target region and OAR is influence the quality of treatment planning. Most cancer patients’radiation therapy is multi-fractional, the image registration technology is always used to optimize the treatment planning and accumulate the dose. Compared with 3D-CRT, IMRT has the better 3D dose distribution,more conformal with PTV and uniformity dose intensity in PTV.Furthermore, the dose decreases quickly outside the PTV and the normal tissue and OARs can be protected efficiently in IMRT. No matter in theory or in clinic, IMRT can complete better in the maximizing the gain ratio of radiation therapy, increase the tumor control probability (TCP), reduce the normal tissue complication probability (NTCP) and improve the quality of patients’life. However, the set-up error and the movement and deformation of tissue and organs between fractions may result in the dose insufficient in target region or normal tissue and organs getting over dose, which may lead to the tumor not controlled and normal tissue and organs damaged, especially in the high dose gradient condition. The main solution is obtaining the deformation information through the registration of the anatomical image in multiple fractions via image registration algorithm, to redesign the initial treatment planning, and then treat the patient with the obtained new planning, which can improve the accuracy of dose patient got greatly. Moreover, the complications after therapy are another important indicator of the therapy effect. The radiation therapy complications are results from the OARs dose exceed the tolerated dose that could lead to dysfunction. The accumulated dose suffered by OARs of patient receive multi-fractional radiation therapy is an important clinic concerned problem. Because of the set-up error and the movement and deformation of tissue and organs between fractions, the dose accumulation can’t be done directly. Therefore, we need to do the OAR registration in each fraction for the accurate dose accumulation.In clinic, the contours of the target region and OARs are main manual segmented by physician. Because the segmentation is relative with the personal knowledge, experience, time, energy etc, the segmentation results may be different from different physicians or in different time by the same physician. The difference would influence the treatment planning accuracy. Moreover, the manual segmentation is time-consumed and low efficiency. The automatic segmentation is all rely on the segmentation algorithm, obtain the result automatic. The automatic segmentation is fast bust low accuracy in segmentation result. Interactive segmentation, which allows the physician to incorporate their professional knowledge and the specific clinical criteria, and saves physicians’ time with computer aid, can obtain optimal segmentation result fast. It becomes the research focus. The other problem is that the target region of complex medical image is always be inhomogeneous. Most segmentation algorithms are based on the intensity similarity. When the intensities of target region are various, common segmentation algorithm would segment the target region as different regions. It is hard to obtain robust result even for the interactive segmentation.In the radiation therapy for gynecology cancer patient, such as cervical cancer patient, intracavitary therapy is the most common treatment method. The intracavitary therapy is high dose-rate brachytherapy. The source is low activity, but the therapy distance is short in intracavitary therapy. The source irradiates the target region directly. Therefore, the target region can receive high dose irradiation, which is effective to improve the TCP. In the procedure of HDR BT, firstly, an applicator was inserted in a suitable position of patient’s vagina; Based on the therapy planning, the residence position and timen was computed for a good dose distribution; Then, the tumor region directly was irradiated through transporting the source to the set position at the top of applicator through the inner tube automatically; Because of the high dose in each fraction, the tumor volume and OARs should be defined explicitly to ensure the minimum dose in tumor volume and control the dose in normal tissue and OARs. In HDR BT, multi-fractions technology is always used. Therefore, a CT image with applicator inserted in the vagina is usually acquired prior for treatment planning purposes (as planning CT). Before each treatment fraction, a similar CT image is also obtained (as HDR therapy CT). The image registration between planning CT and HDR CT can be used to correct the set up error, redesign the treatment planning, or accumulate dose for treatment planning verification. In HDR BT, one patient may receive different type of radiation therapy in different fractions, such as HDR and IMRT.In addition, different applicators could be used in different HDR fractions. There are many types of applicator, such as, cylinder applicator, tandem & oval (T&O) applicator and tandem & cylinder (T&C) applicator, et al.At present, there are still some open problems in the registration between HDR CT images and the OAR dose accumulation in HDR BT.Firstly, because the information is dissymmetry between HDR CT images in different fractions, general image registration methods failed on HDR CT image registration. Almost all image registration algorithms are based on the assumption that the anatomical structure in the two images must be point-to-point correspondence, which means that a point can be found on an image corresponding to the point on the other. However, the registration between HDR CT images with applicator is against to this assumption. There is a nonexistent or dissymmetry structure (applicator) on planning CT which is existent on HDR CT. Therefore, the general image registration can’t solve this registration problem.Secondly, most of the image registration algorithms perform poor on low contrast region. In the HDR BT for gynecology cancer patient, the OAR (bladder, rectum, etc) may be involved in the high dose region. A quarter of the bladder wall would be involved in the high dose region in each HDR BT. In addition, once the bladder wall is damaged by radiation, the whole bladder would have dysfunction. What is worse, the bladder and rectum always show low contrast characteristic in CT image. If general image registration algorithm is used for the OAR registration, the poor registration result may introduce large error to the dose accumulation, lead to inaccuracy accumulated dose.This research focus on the image segmentation and the image registration for HDR BT, aims to solve the inhomogeneous image segmentation problem, the registration of HDR CT images and bladder surface accurate registration. This research mainly contains:(1) We propose a Seed Points Auto-generation for RW Segmentation Enhancement (SPARSE) algorithm. The segmentation results of synthetic phantom image and clinical patients’ images Demonstrate the SPARSE algorithm is robust and accuracy. We applied the SPRSE algorithm for the inhomogeneous target segmentation of morphological MR and CT images. We make use of the intensity information of target and background contained in the initial seed points labeled by physician, propose a seed points auto-generation method which growing based on the initial seed points to extended seed points. Then, appling RW based on the extended seed points, we can get more accuracy segmentation result. The improved method can decrease the influence from seed points’ quantity and location to RW, and weak influence of the choice of parameter to RW.(2) We propose a Segmentation and Point matching Enhanced Efficient DIR algorithm (SPEED) to solve the problem if information asymmetrical in HDR CT image. We demonstrated the effectiveness and evaluated the accuracy of SPEED using the synthetic data and clinical data. Firstly, we segment the applicator region in moving and reference image, and fill it with air’s CT value. Then we make the air cave surface points matching through the thin plate spline robust point matching algorithm(TPS-RPM), and approximate the deformation vector field(DVFs) via B-Spline method. Therefore, we can achieve the whole image registration by applying the Demons algorithm with the approximated DVFs as the initial DVFs. The SPEED solved the HDR CT image registration and improved the registration accuracy in the applicator region largely.(3) To achieve the bladder surface dose accumulation, we propose an improved TPS-RPM method with local topology preservation (named TPS-RPM-LTP). This method is based on the TPS-RPM, but considers the characteristic that the relationship of the anatomical points and their surrounding points is preserved when the tissue or organs deformed. Therefore, we add the constrain of local topology preservation in each iteration.In addition, we used a homemade bladder phantom, two simulated cases and clinical data from seven patients (twenty nine fractions)to Demonstrate the accuracy of TPS-RPM-LTP. The results show that the TPS-RPM-LTP can improve the accuracy of bladder surface matching, which is hopeful to be used for bladder surface dose accumulation.