Monte Carlo Framework for Prostate Cancer Correction and Reconstruction in Endorectal Multi-parametric MRI

Book excerpt: Prostate cancer is one of the leading causes of cancer death in the male population. The detection of prostate cancer using imaging has been challenging until recently. Multi-parametric MRI has been shown to allow accurate localization of the cancers and can help direct biopsies to cancer foci which is required to plan treatment. The interpretation of MRI, however, requires a high level of expertise and review of large multi-parametric data sets. An endorectal receiver coil is often used to improve signal-to-noise ratio (SNR) and aid in detection of smaller cancer foci. Despite increased SNR, intensity bias fields can exist where nearest the endorectal coil the signal is greater than those regions farther from the coil. Weak delineation of the prostate as well as poor prostate gland visualization can greatly impact the ease and accuracy of diagnosis. For this reason, there is a need for an automated system which can correct endorectal multi-parametric MRI for enhanced visualization. A framework using Monte Carlo sampling techniques has been developed for prostate cancer correction and reconstruction in endorectal multi-parametric MRI. Its performance against state-of-the-art approaches demonstrate improved results for visualization and prostate delineation. The first step in the proposed framework involves reconstructing an intensity bias-free image. Using importance-weighted Monte Carlo sampling, the intensity bias field is estimated to approximate the bias-free result. However, the reconstruction is still pervaded by noise which becomes amplified and non-stationary as a result of intensity bias correction. The second step in the framework applies a spatially-adaptive Rician distributed Monte Carlo sampling approach while accounting for the endorectal coil's underlying SNR characteristics. To evaluate the framework, the individual steps are compared against state-of-the-art approaches using phantoms and real patient data to quantify visualization improvement. The intensity bias correction technique is critiqued based on detail preservation and delineation of the prostate from the background as well as improvement in tumor identification. The noise compensation approach is considered based on the noise suppression, contrast of tissue as well as preservation of details and texture. Utilizing quantitative and qualitative metrics in addition to visual analysis, the experimental results demonstrated that the proposed framework allows for improved visualization, with increased delineation of the prostate and preservation of tissue textures and details. This allows radiologists to more easily identify characteristics of cancerous and healthy tissue leading to more accurate and confident diagnoses.