Ulas Bagci PhD, MSc, Presenter: Nothing to Disclose

Jayaram K. Udupa PhD, Abstract Co-Author: Nothing to Disclose

Neil Mendhiratta, Abstract Co-Author: Nothing to Disclose

Kirsten Jaster-Miller BEng, Abstract Co-Author: Nothing to Disclose

Daniel Joseph Mollura MD, Abstract Co-Author: Nothing to Disclose

To develop a simultaneous segmentation method for automatically delineating regions of radiotracer uptake and corresponding anatomical structures in MRI-PET-scans for precise quantitative analysis.

We retrospectively analyzed 10 MRI-PET scans (Siemens Biograph mMR system) of patients with renal cell carcinoma and metastatic colon cancer. The proposed automatic segmentation method includes two key steps: (1) fast localization of high uptake regions (see Fig 1a), and (2) novel simultaneous delineation method based on random walk boundary detection (Fig 1b and c (zoomed b) where segmented object (white boundary) and ground truth (yellow boundary) are shown). Prior to these key steps, we co-registered PET images to MRI in order to avoid any uptake localization error due to possible misalignments. We identified radiotracer uptake regions by computing 40% of the SUVmax value from PET images. We found non-object areas by searching pixels having less than 40% of SUVmax in 8-connectivity of foreground seeds. Next, our algorithm utilizes information from foreground and background seeds and finds the optimal boundary using textural properties of both PET and MRI images. We used dice similarity coefficient (DSC) and Hausdorff distance (HD) to evaluate segmentation accuracy. Higher values of DSC and lower value of HDs indicate more accurate segmentations.

For the registration, we used fast affine registration method with 20K iterations at most. We used nearly 700K spatial samples from each modality to compute similarity metric over the pairwise mutual information histogram. The whole registration process takes an avg of 13 sec. While inter-observer agreement between two experts in segmenting abnormal activity regions was reported as DSC of 83.24% with 7 mm HD, intra-observer agreement was reported as DSC of 92.63% with 2.23 mm HD. The proposed segmentation method achieved a DSC of 86.91% with HD of 5.09 mm. To the best of our knowledge, this is the first study showing automated segmentation of mMR images simultaneously.

The preliminary study showed the feasibility of a fast, accurate, and automated method for simultaneously segmenting radiotracer uptake regions in MRI-PET images.

Accurately assessing radiotracer uptake from PET with simultaneous corresponding anatomical tissue boundaries from MRI via automated optimized segmentation is an important application in computational

Bagci, U, Udupa, J, Mendhiratta, N, Jaster-Miller, K, Mollura, D, Simultaneous Segmentation of Abnormal Activities from Hybrid MRI-PET Images.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12027747.html