Transcript Slide 1
Visualization and Detection of Prostatic Carcinoma Joseph Marino, Xin Zhao, Ruirui Jiang, Wei Zeng, Arie Kaufman, Xianfeng Gu Stony Brook University, Stony Brook, NY 11794-4400 Introduction Visualization & Detection Registration • Prostate cancer is the most commonly diagnosed cancer and the second leading cause of cancerrelated mortality among U.S. males • Detection is inexact, relying on PSA blood tests, digital rectal examinations, and multiple biopsies that are generally stabbing in the dark • MRI has been suggested as an imaging modality to help locate prostate cancer • Computer analysis techniques can make this task easier and more exact • Segmentation of only one dataset is needed • Rendering performed via volumetric ray casting • At each sample point within the peripheral zone, a score is calculated using the six modes: • Registration for scans acquired at different times or patient positions (not naturally registered) • Feature points are needed to align the datasets and can be found using corner detection: Data • Multiple orientations and modes of MR data are acquired for prostate cancer detection • We use five image sequences for each patient: T2 Axial T2 Sagittal MRSI T2 Coronal Score = MRSIA + MRSIB + T2A + T2S + T2C + T1A Where: MRSIA = (ratioA – threshMRSI) x percentage x 0.5 MRSIB = (ratioB – threshMRSI) x percentage x 0.5 T2A = (threshT2 – T2axial) x 0.333 T2S = (threshT2 – T2sagittal) x 0.333 T2C = (threshT2 – T2coronal) x 0.333 T1A = threshT1 – T1axial This scoring corresponds to the following: A higher MRSI ratio indicates cancer Lower intensity T2 areas indicate cancer Higher intensity T1 areas indicate not cancer • Positive score indicates likelihood for cancer. • Scoring is integrated into the visualization of the prostatic volume • Areas of high likelihood for cancer are mapped to red, and low likelihood areas are mapped to blue: T1 Axial • The position & orientation information are known for each image sequence • They can be aligned with respect to each other: Cancer indicated in left & right midgland & base Segmented Prostate Slice Edges & Corners Final Detected Features • Map the two prostate volumes to balls using volumetric conformal mapping • Align the volumes using the feature points: MR Prostate Histology Unaligned Aligned Proposed Work • Continue to explore better methods of detecting cancer and registering different scans • Investigate further MR modalities which can improve cancer detection (e.g., diffusionweighted, perfusion, dynamic contrast enhanced) • As more modalities are introduced into our framework, the data becomes greater and we strive to handle it all in an interactive manner • Interventional visualization and detection as patients are in the scanner in order to localize treatment delivery • Clinical studies to determine optimal analysis parameters and user friendliness