Nikolaos Karakatsanis Laboratory

In terms of PET data acquisition protocols, the lab is designing clinically adoptable dynamic whole-body human PET scan protocols customized for the short AFOV human PET scanners that are widely available today. The aim is to utilize those imaging protocols to translate the quantitative virtues of truly multi-parametric, beyond static SUV PET imaging to oncology, cardiology, and multi-organ (e.g., brain-heart axis) studies. As a pioneer in the development of the first dynamic whole-body 18 fluorodeoxyglycose (FDG) PET/CT human scan protocols a few years ago, the lab now collaborates closely with nuclear medicine and radiology physicians to expand and customize those PET protocols for other important radiotracers such as 68 gallium (Ga)-prostate specific membrane antigen (PSMA) and 68Ga-DOTATATE to benefit diagnosis and treatment response assessment for metastatic castrate-resistant prostate cancer (mCRPC), neuroendocrine tumors (NETs), and cardiovascular diseases. The lab is investigating novel dual-radiotracer PET/MR acquisition protocols, with 18F-sodium fluoride (NaF) being one of the two administered radiotracers acting as an agent of unique dynamics to drive bone segmentation and aid in the synthesis of more complete tissue attenuation maps. The result: more accurate attenuation correction of the detected PET signal in PET/MR studies where CT modality, the gold standard for PET attenuation correction, is missing.  

Regarding PET image reconstruction, the lab introduced novel direct 4D PET image reconstruction algorithms tailored for multi-bed or continuous bed motion acquisitions to support multi-parametric whole-body PET imaging even with the limited AFOV clinical PET scanners widely accessible now. In those algorithms, the lab incorporated robust graphical analysis methods enabling direct imaging of multiple clinically relevant macro-kinetic parameters of the administered radiotracers with high precision and accuracy despite the high noise levels typically present in the dynamic whole-body raw PET data. Aside from their robustness, the utilized graphical analysis methods can significantly shorten the total scan period required to capture the macro-dynamics of the administered radiotracer. The lab aims to match the scan periods post-injection (p.i.) that are typically used for routine standard-of-care static PET scan protocols, e.g., 60-80 min p.i. in the case of whole-body 18F-FDG PET exams, without compromising for kinetic macro-parameters accuracy and precision. The lab is collecting independent measurements of specific radiotracers’ concentration in arterial blood plasma, known as input function, from respective populations of past dynamic PET exams to build population-based models for each specific radiotracer. These population-based models let the lab acquire prospectively only a small portion of the input function at later and shorter time windows post-administration, matching the time window employed for typical state-of-the-art static PET scans, allowing for inference of the missing earlier section from the population-based model. 

The lab also investigates novel methods of data normalization and machine-learning (ML) aided image reconstruction customized for PET data acquired with cost-effective sparse PET detector geometries with double the AFOV of the respective compact PET detector geometries. To attain high image quality for reconstructed sparse PET data, similar to that attained with compact PET scanners, the lab employed a customized component-based normalization method accounting for the sparse sampling effects due to gaps presence. The lab developed a novel limited continuous bed motion (limited-CBM) PET acquisition mode for the sparse geometries to smooth the expected large variance in axial sensitivity observed between the large gaps and the physical detector blocks. The lab is also investigating, with other Cornell artificial intelligence experts, ML approaches based on U-net deep convolutional neural networks (CNNs) to recover losses in image quality and mitigate elevated noise levels due to sparse sampling reconstructing and analyzing sparse PET data. The lab sees this as very important as it can allow the development of cost-effective clinical PET scanners of adaptable AFOVs supporting, under a single universal scanner architecture, multiple customized configurations optimized for the unique demands of each research or clinical PET study. Future generation clinical PET scanners may adapt their configuration optimally to offer either highly sensitive scans over shorter AFOVs, or moderately sensitive scans over longer AFOVs, depending on the targeted imaging task of each study.