For newly diagnosed prostate cancer patients with a positive biopsy, there are a variety of treatment options to consider. To aid physicians and patients in their decision making, a variety of predictive assays have emerged within the last decade, many of them imaging based. These assays build predictive models for survival analysis to provide personalized risk assessments for the patients. However, there have rarely been any published studies on how the amount of tumor in the positive prostate biopsy affects the predictive power of these imaging based assays. Recently we have proposed a new algorithmic framework for survival analysis employing semi-supervised transductive regression. This approach has improved the predictive power of biopsy based prostate cancer prognostic models. In this paper, we explore how different amounts of tumor in the prostate biopsy affect the accuracy of imaging based prognostic models employing this framework. We show that the framework improves accuracy even with diminishing amounts of tumor, thereby enabling more accurate treatment decisions.