Our overall objective is to develop numerical observers for dependable technology evaluations in emission tomography. Positron emission tomography (PET) and single-photon emission tomography (SPECT) are the primary clinical modalities for imaging many types of cancer. However, early de- tection often presents the best chances of surviving cancer whereas these imaging modalities have limited diagnostic utility for small tumors. Systematic task-based developmental assessments could facilitate early identification of promising new technology for improving the diagnostic capabilities of these modalities. Yet, assessments with human observers are generally impractical for developmental use. Moreover, available mathematical models (or numerical observers) intended to predict human performance-what we refer to as human-model observers-present significant limits, including con- straints on the types of diagnostic tasks that can be considered. Partly because of these constraints, existing human-model observers frequently require revalidation given any change to the imaging pro- cess. Our approach to observer development is founded on the concept of task equivalence, whereby the task for the numerical observer mirrors the desired clinical task as closely as possible. In this grant, we propose a novel observer framework that is influenced by descriptions of radiologists' visual-search (VS) processes, in which an initial global scan of an image identifies candidate locations deserving closer inspection. We shall use the VS paradigm to investigate the hypotheses that task equivalence i) can lead to a human-observer model that reliably generalizes to a wide range of diagnostic tasks, and ii) is necessary to ensure truly relevant task-based evaluations. We shall test these hypotheses through observer studies with fluorine-18 deoxyglucose (FDG) whole-body PET and SPECT In-111 imaging of neuroendocrine tumors (NETs). A state-of-the-art model observer for developmental stud- ies should be capable of detection-localization tasks, and our observer studies will be analyzed with jackknife FROC (JAFROC) methodology.
The specific aims of this work are to: 1) determine what features of FDG-PET image slices attract initial human-observer attention; 2) develop a VS numerical observer for tumor detection-localization tasks in FDG-PET; 3) test the VS observer against humans in a JAFROC detection-localization study featuring hybrid PET images; 4) investigate generalizations of the VS observer to SPECT and 3D detection-localization tasks; and 5) compare system optimiza- tions for oncologic SPECT obtained from the VS and existing numerical observers. The application is optimization of a parallel-hole collimator design for In-111 NET imaging.
Functional imaging with positron emission tomography (PET) and single-photon emission tomog- raphy (SPECT) has a significant role in cancer diagnosis and management. Still, early detection offers the best chances for surviving many cancers and refining the diagnostic capabilities of these modali- ties for small tumors continues to be a major research focus. This work is focused on the development of assessment methods for assisting in the early identification of technological advances that could improve the diagnostic utility of PET and SPECT.
|Gifford, Howard C; Liang, Zhihua; Das, Mini (2016) Visual-search observers for assessing tomographic x-ray image quality. Med Phys 43:1563-75|
|Sen, Anando; Kalantari, Faraz; Gifford, Howard C (2016) Task Equivalence for Model and Human-Observer Comparisons in SPECT Localization Studies. IEEE Trans Nucl Sci 63:1426-1434|
|Gifford, H C (2014) Efficient visual-search model observers for PET. Br J Radiol 87:20140017|
|Lau, Beverly A; Das, Mini; Gifford, Howard C (2013) Towards Visual-Search Model Observers for Mass Detection in Breast Tomosynthesis. Proc SPIE Int Soc Opt Eng 8668:|
|Gifford, Howard C (2013) A visual-search model observer for multislice-multiview SPECT images. Med Phys 40:092505|
|Das, Mini; Gifford, Howard C; O'Connor, J Michael et al. (2011) Penalized maximum likelihood reconstruction for improved microcalcification detection in breast tomosynthesis. IEEE Trans Med Imaging 30:904-14|