Supervised prostate cancer segmentation with multispectral MRI incorporating location information

Prostate cancer is considered to be one of the main causes of cancer related death for men in the United States. Automated methods for prostate cancer localization based on multispectral magnetic resonance imaging (MRI) haver recently emerged as a non invasive technique for this purpose as an alternative to transrectal ultrasound. However, the automated methods developed to this date require a manual segmentation of the peripheral zone (PZ) of the prostate. This paper proposes a supervised method that removes the need for PZ extraction based on support vector machines (SVM) that considers location information in addition to the intensity values of the multispectral MRI in the classifier. In this way, subjective and inefficient manual PZ extraction is eliminated. We demonstrate the effectiveness of the algorithm by applying it to multispectral MRI data from 21 biopsy confirmed cancer patients and providing both quantitative and visual results.