Prostate Segmentation From 3 DUS Using Regional Texture Classification and Shape Differences-Late Draft

Accurate diagnosis of prostate cancer relies on an accurate biopsy procedure. The biopsy must be taken from the center of the suspected malignant region in order to reliably diagnose the patient’s condition. Tissue taken from the edge of this region, or from outside this region entirely, can result in either a misdiagnosis of the type of cancer or a fully negative test, either of which could prove harmful to the patient. When planning a biopsy, a physician will inspect the prostate using an imaging modality such as MRI in order to identify regions where cancer is believed to be present. This planning image, however, is not what the physician performing the biopsy will see. A biopsy-time imaging modality must be non-ionizing and able to be performed in clinically useful time. The modality used is ultrasound, which sacrifices the quality of MRI for speed and ease of use. During the biopsy procedure, the biopsy needle/ultrasound transducer assembly must be inserted into the rectum in order to most effectively reach the prostate. Due to the proximity of the prostate to the rectum, this insertion causes deformation in the prostate and surrounding anatomy. These deformations, combined with the lower quality biopsy imaging, make accurately locating the regions of interest identified on the MRI image difficult. In order to transfer this information from the planning-time MRI image to the biopsy-time TRUS image, two major challenges must be overcome. The first of these is to account for the deformations caused by the insertion of the biopsy needle and ultrasound transudcer in order to form a reliable correspondence between the prostate in between planning-time and biopsy-time. The second is to use this deformation knowledge to locate and segment the prostate in ultrasound, which is made difficult due to the similarity in appearance of the prostate to some of its surrounding anatomy in an ultrasound image. In order to segment the prostate in ultrasound, we introduce a method based on voxel classification. Due to the varying appearance of the anatomy surrounding the prostate, a regional approach is taken to ensure these areas with different