Search

Clinical History: A previously healthy 47-year-old male presented with a 1-week history of fever, vomiting, nausea, and epigastric pain. Laboratory studies were notable for lymphopenia (0.8×103/µL, normal range 0.9×103/µL – 3.3×103/µL). Vitals signs were remarkable only for low-grade fever (37.9°C). He reported no respiratory complaints or known sick contacts., 47 years, male

Prostate cancer is the most diagnosed cancer among men. Many patients with localized prostate cancer are treated with brachytherapy, one form of which involves permanent implantation of approximately 100 radioactive sources into and sometimes immediately around the prostate while the patient is anesthetized. During the procedure, transrectal ultrasound (TRUS) and fluoroscopic images (acquired using a mobile C-arm fluoroscopic X-ray system) are used to guide and visually assess implant quality, but do not provide accurate quantitative dosimetry. Thus, the patient undergoes a CT scan after the implantation is completed for dosimetric evaluation. However, this practice is not ideal as it occurs after the patient has left the operating room, when there is no longer any opportunity to modify the implant, if required. In this research project, a workflow was developed to assess the feasibility of performing intraoperative dosimetry using two routinely available imaging systems (a cone beam CT (CBCT) capable C-arm, and an ultrasound machine) for intraoperative dosimetric assessment of permanent implant prostate brachytherapy. In the proposed methods, the locations of all implanted sources were obtained from either 3D reconstructions of multiple planar radiographs, or from CBCT images. They were then registered to prostate contours delineated on the TRUS images, based on a common subset of sources identified on both image sets. In this process, prostate contours were deformed, using a finite element model, to take into account the effect of probe pressure in the TRUS images. Prostate dosimetric parameters obtained using this method were in agreement with postimplant CT dosimetry results, considering the uncertainty associated with each of these methods. An algorithm for automatic detection of seeds on TRUS images using a convolutional neural network was also developed during the course of this work. The model was trained to detect the needle track first and then the individual sources within the needle track. This automated approach outperformed a human observer in precision. The results of the work described in this thesis support the conclusion that the proposed dosimetry methods are feasible for real-time intraoperative dosimetric analysis of the implant and can potentially also replace postimplant CT dosimetry.