| Bone scintigraphy is a useful tool for tumor detection and bone metastasis analysis in nuclear medicine. It has been widely used because of high sensitivity, whole-body imaging abilities and relatively low price. However, the differences between patients and physicians'experience, the poor image quality have limited the accuracy of image interpretation. So it is meaningful to develop computer aided diagnosis (CAD) system for whole body bone scintigraphy, especially for the quantitive analysis of tumor and patients. Image segmentation and registration are both key issues for CAD system. Besides, the low signal noise ratio (SNR), the weak boundary and the varieties among patients all pose new theoretical challenges.In this paper, the author takes throughout researches in image partition, segmentation and registration based on the characteristics of bone scintigraphy. Several adaptive algorithms have been proposed in practical use for the CAD system. Firstly, an adaptive image partition method is proposed based on the analysis of normal patients. Combined with prior knowledge about spine, a new scheme for spine segmentation is proposed, which can deal with the shape and intensity varieties among different patients. Using the results of thoracic area and spine, the author proposes a new algorithm for rib segmentation based on the extraction of rib skeleton. Experiments show that this template and prior knowledge based method can generate more satisfactory results than existing methods. The segmentation algorithms have been used in the CAD system, which can meet the clinical needs.Besides, in this paper the image registration problem for bone scintigraphy has been discussed. Several registration algorithms are discussed and the Demon algorithm is chosen in this application. This optical flow based method is applied for image alignment and hotspot detection as well. The CAD system based on these algorithms has been tested clinically and shows promising prospects. |
PDF Full Text Request