Molecular imaging of the human body is beginning to revolutionize drug development, drug delivery targeting, prognostics and diagnostics, and patient screening for clinical trials. The primary clinical tool of molecular imaging is Positron Emission Tomography (PET), which uses radioactively tagged probes (radioligands) for the in vivo quantification of blood flow, metabolism, protein distribution, gene expression and drug target occupancy. While many radioligands are used in human research, only a few have been adopted for clinical use. A major obstacle to translating these tools from bench-to-bedside is that PET images acquired using complex radioligands can not be properly interpreted or quantified without arterial blood sampling during the scan. Arterial blood sampling is an invasive, risky, costly, time consuming and uncomfortable procedure that deters subjects' participation and requires highly specialized medical staff presence and laboratories to run blood analysis. Many approaches have been developed over the years to reduce the number of blood samples for certain classes of radioligands, yet the ultimate goal of zero blood samples has remained illusive. In this dissertation we break this proverbial blood barrier and present for the first time a non-invasive PET quantification framework. To accomplish this, we introduce novel image processing, modeling, and tomographic reconstruction tools.
First, we developed dedicated pharmacokinetic modeling, machine learning and optimization framework based on the fusion of Electronic Health Records (EHR) data with dynamic PET brain imaging information. EHR data is used to infer individualized metabolism and clearance rates of the radioligand from the body. This is combined with simultaneous estimation on multiple distinct regions of the PET image. A substantial part of this effort involved curating, and then mining, an extensive database of PET, EHR and arterial blood sampling data.
Second, we outline a new tomographic reconstruction and resolution modeling approach that takes into account the scanner point spread function in order to improve the resolution of existing PET data-sets. This technique allows visualization and quantification of structures smaller than previously possible. Recovery of signal from blood vessels and integration with the non-invasive framework is demonstrated. We also show general applicability of this technique for visualization and signal recovery from the raphe, a sub-resolution cluster of nuclei in the brain that were previously not detectible with standard techniques.
Our framework can be generalizable to all classes of radioligands, independent of their kinetics and distribution within body. Work presented in this thesis will allow the PET scientific and clinical community to advance towards the ultimate goal of making PET cost-effective and to enable new clinical use cases.
Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8222SQ1 |
Date | January 2015 |
Creators | Mikhno, Arthur |
Source Sets | Columbia University |
Language | English |
Detected Language | English |
Type | Theses |
Page generated in 6.7627 seconds