Research On Needle Detection In 3D Ultrasound Images

Ultrasound(US) image-guided interventional therapy is one of the most important branches in the area of medical ultrasound. It mainly includes two parts: ultrasound image-guided biopsy for lesion characteristic identification; ultrasound image-guided operation of lesions therapy and drug delivery. Compared with traditional 2D ultrasound image, 3D ultrasound image has the following advantages for image-guided therapy: it can help the surgeons to observe the structures and shapes of the objects from different views; it can also demonstrate the distances clearly in 3D space between organs and medical intervention instruments, so that the possible accidents can be avoided. 3D ultrasound image-guided intervention operation has been one of the main streams in medical ultrasound image development.One of the essential technology in 3D US image-guided interventional operation is to locate the interventional instrument accurately when the 3D US images of the lesions are received, so as to help the surgeons to reduce the amount of images needed to read and improve the efficiency of reading images. Aiming at the difficulty of locating the Radio Frequency(RF) electrode in the treatment of uterine myoma using 3D US image-guided RF interventional operation, an automatical needle detection based on 3D Hough transform in 3D US image was proposed. This thesis mainly focuses on the extension of 2D Hough transform to 3D Hough transform which is applied in the needle detection of 3D US images, it includes:(1)A real time 2D gray-level Hough transform was proposed which combines the phase grouping and traditional 2D gray-level Hough transform. Compared with the traditional 2D Hough transform, it does not need the binarization and its processing speed is faster. Experiment results showed that the method can satisfy the requirement of the image-guided surgery with angular error lower than 1°and position error less than 0.5mm.(2) 2D Hough transform was extended to 3D space. Several line representations in 3D space, such as standard point-vector representation, 4-parameter point-vector representation, Denavit—Hartenberg line representation, Roberts line representation and Plücker line presentation, were presented. Each line representation's corresponding 3D Hough transform was proposed. Considering the mathematical elegancy, computational complexity and mem- ory requirements, the standard 3D Hough transform based on the Roberts line representation was established, which is called 3D standard Hough transform(3DSHT). The experimental results showed that 3DSHT can detect the line accurately in the 3D synthetic images and 3D synthetic noisy images.(3) An automatical needle detection algorithm based on the 3DSHT was developed in 3D US images. The experimental results showed that our method worked well in the 3D US images of water phantom, agar phantom and chicken phantom. The proposed method was also verified with different cropping size and insertion depth of needle(4) Several fast algorithms based on standard 3D Hough transform were proposed to speed up the 3DSHT: 3D Hough transform based on the coarse-fine search strategy, 3D Randomized Hough transform, and 3D Randomized Hough transform based on the coarse-fine search strategy. Compared with the experimental result of each algorithm, 3D Randomized Hough transform based on the coarse-fine search strategy is fastest among the three algorithms.