Evaluating treatment related changes for prostate cancer via image analysis tools and magnetic resonance imaging.

摘要

The goal of this work is to quantitatively evaluate treatment response following prostate cancer treatment, via the development of novel segmentation and registration methods for: radical prostatectomy, focal laser ablation (FLA) and external beam radiation treatment (EBRT) imaging data. Radical prostatectomy specimens are evaluated via accurately quantifying the prostate volume pre- and post-treatment. A novel Multi-Feature, Landmark Free Active Appearance Model (MFLAAM) algorithm has been developed to automatically determine the volume. This is compared to submersing the removed prostate in water. Quantitative results show that the MFLAAM yields more accurate segmentations than existing state of the art systems, and offers highly accurate volume estimations compared to current clinical volume estimation procedures. In order to evaluate EBRT and FLA treatments for prostate cancer, the pre-, post-treatment MRI images must be spatially aligned. However, existing tools do not take into account specific treatment related changes to the prostate. In addition, no automatic quantitative tools for specifically evaluating treatment changes exist. The prostate consists of distinct internal substructures central gland (CG) and peripheral zone (PZ). Our model aims to explicitly take into account the different effects treatment may have on the shapes of the internal prostatic structures. In order to automatically segment the CG and PZ, the MFLAAM algorithm was extended to simultaneously segment multiple objects. Following the automatic segmentation, a finite element model (FEM) registration algorithm is introduced. An FEM uses physical properties to constrain the registration to only physically-real deformations. In addition, the shrinking of the prostate (which occurs due to radiation treatment) is specifically modeled. This FEM was quantitatively compared to other registration techniques, and was the best performing algorithm over 30 patients. Finally, a separate FEM is developed to compensate for the changes in the surrounding organs (bladder and rectum filling), which is essential for one to isolate the treatment-related changes in the prostate. Following an accurate registration, changes in the MR parameters, changes in the prostate volume, and changes in prostate morphology are calculated. We envision that this work will pave the way for predictive models in order to predict patient outcome from early follow-up imaging data.