Image-Guided Targeting In Treatment Planning For Focused Ultrasound Therapy

High intensity focused ultrasound (HIFU) is one of the advancing non-invasive surgical tools for treating diseases, such as tumors. Treatment planning, especially for 3D targeting, is one of the key elements affecting its therapeutic efficacy. The purpose of image-guided technology is to use medical imaging devices to perform the task of targeting. Among all the devices, B mode ultrasound and magnetic resonance (MR) are most popular in clinic. B mode ultrasound has the advantages of lower cost, accessibility, and faster scaning time. However, poor image quality makes it difficult for accurate targeting. On the contrary, MR has the benefits of better image quality and multiparametric imaging as well as the ability to monitor temperature. However, it is much more expensive and complex to be integrated with therapeutic transducers.To address the difficult problem of targeting in ultrasound images, this paper uses preoperative CT/MR images and real time B mode ultrasound to do the targeting together. The benefit is realizing accurate targeting by CT/MR images without the necessity of installing MR in HIFU. The guiding procedure involves image acquisition, preprocessing, segmentation, reconstruction of target regions, and multi-modality image registration.In image preprocessing, a new model called stick-guided lateral inhibition is proposed to enhance small line structures without losing useful information of original images. The model has the ability to inhibit the synchronic amplication of noises during enhancing interested structures. Hence, important structures are effectively enhanced and feature extraction in latter registration is simplified. To accelerate the denoising speed by diffusion stick model, we propose to sample stick filtering kernels by equal interval. While significantly reducing computing time, the proposed method has satisfying results as the original model.In image segmentation, with certain initialization strategies in level set method and active contour model, we develop fast semiautomatic and completely automatic segmentation approaches. A new algorithm based on Cartoon-Texture model is presented for automatic tracing and segmentation of target regions in image sequences, which does not require that the number of target regions be constant and spatial location of target regions be adjacent among different image slices. It can also remove some adjoining tissues attached to the regions as well as solving the problem of over-segmentation of small holes inside the regions.In reconstruction of target regions, a self-adaptive 3D surface reconstruction algorithm based on the shortest adjacent slices distance is proposed. Of simple implementation, it can quickly reconstruct target regions from segmented image slices. By 3D visualizing focused points inside the reconstructed surface, users can directly adjust and optimize the treatment plan. Therefore, the accuracy of treatment planning is improved.In image registration, multi-modality image registration and fusion based on Tri-Views is proposed, which greatly simplifies the extraction of 3D corresponding points for rigid registration and parameters setting for non-rigid registration. After registration, we are able to map the target regions and focused points from CT/MR image space to ultrasound image space. Hence, accurate targeting in ultrasound images is achieved.