Georgios Z. Papadakis MD, Presenter: Nothing to Disclose

Ulas Bagci PhD, MSc, Abstract Co-Author: Nothing to Disclose

Brent Foster, Abstract Co-Author: Nothing to Disclose

Ziyue Xu PhD, Abstract Co-Author: Nothing to Disclose

Awais Mansoor PhD, Abstract Co-Author: Nothing to Disclose

Nicholas John Patronas MD, Abstract Co-Author: Nothing to Disclose

Constantine Argyrios Stratakis MD, DSc, Abstract Co-Author: Nothing to Disclose

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

(1) To present an automated computational tool for accurate and fast image analysis of PET-CT scans, and (2) to compare computer-derived imaging markers of various biopsy proven lesions with radiologists’ manual assessment.

With IRB approval, we retrospectively analyzed PET-CT images of 15 patients who were diagnosed with granulomatous inflammation, cystadenomas, neurofibromas, neuroendocrine tumors, renal cell carcinomas, several types of lymphomas, and a case of breast carcinoma. All lesions were biopsy proven. Our proposed framework contains a PET image segmentation method which is based on the affinity propagation based clustering algorithm. Our optimal segmentation algorithm segments focal and multi-focal lesions within a few seconds in 3-D image space. We compared the computer derived SUV-based statistics and metabolic tumor volumes with radiologists-derived measures to test the feasibility of using this software for PET image analysis for clinical practice.

We obtained a strong correlation (R > 0.91, p< 0.0001) between the SUVmax measurements derived automatically from our framework and the radiologist-derived measurements. We also observed that the radiologist-derived region of interest (ROI) for metabolic tumor volume assessment was significantly over-estimated, and did not show the correct boundary of the lesions. Furthermore, when radiologists used conventional thresholding based delineation algorithms, there were additional false positive removal steps necessary for proper lesion volume assessment. Figure 1 shows a sample view of the radiologist-derived elliptic ROI (left) as well as the true boundary and statistics of the computer-derived results (right).

The proposed automatic tool can be used to derive SUV-based measures as well as metabolic tumor volume in a more accurate and efficient manner. Our proposed framework is open-source, freely available, and will allow researchers to conduct PET quantification studies in routine clinics.

Accurately assessing radiotracer uptake from PET and quantifying the metabolic tumor volume are important precursors in diagnostic decision mechanisms. The presented automated method can be employed for a routine use.

Papadakis, G, Bagci, U, Foster, B, Xu, Z, Mansoor, A, Patronas, N, Stratakis, C, Mollura, D, Automated Computer-derived SUV and Metabolic Tumor Volume Measurements of Biopsy Proven Lesions: Comparison with Radiologist-derived PET-CT Imaging.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14015577.html