Optimizing Dual-Time and Serial Positron Emission Tomography and Single Photon Emission Computed Tomography Scans for Diagnoses and Therapy Monitoring

IntroductionA region’s early and late tracer uptake activities, QE and QL, within a dual-time scan (i.e. using two frames) or in serial scans (as for monitoring therapeutic response), are popular quantitative diagnostic aids, especially in oncology. In this paper, maximum performance is sought from their joint use.Methods$$ {Q_{L} } \mathord{\left/ {\vphantom {{Q_{L} } {Q_{E} ^{n} }}} \right. \kern-\nulldelimiterspace} {Q_{E} ^{n} } $$ is introduced as a tumor marker with an empirical n. This generalizes traditional data weighting having n = 1 for QL/QE, the retention index (RI), with its associated % difference. Using patient data, iterative guessing finds an optimal n that maximizes a measure of diagnostic performance: D = (difference of normal and abnormal marker means)/(their combined SD), which may be computed from values of $$ {Q_{L} } \mathord{\left/ {\vphantom {{Q_{L} } {Q_{E} ^{n} }}} \right. \kern-\nulldelimiterspace} {Q_{E} ^{n} } $$, as well as of QL, QE, and RI each used alone. For 2-deoxy-2-[F-18]fluoro-d-glucose(FDG)-positron emission tomography (PET) dual-time protocols, another approach to optimization—selection of scan times—is investigated by simulations using the Sokolov model.ResultsA meta-analysis of 12 PET and single photon emission computed tomography (SPECT) studies with various tracers, cancers, and scan classes (dual-time or serial) finds ns from 0.5 to 1.1. The optimal D necessarily exceeds the best (or any) computed using QE, QL, or RI: negligibly to by as much as 0.6 (or 1.5). The increases in optimal receiver operating curve area (Az) over the best (or any) traditional marker range from negligible to 0.07 (or 0.4). QE alone usually has the lowest D and Az. Statistically significant performance improvement of $$ {Q_{L} } \mathord{\left/ {\vphantom {{Q_{L} } {Q_{E} ^{n} }}} \right. \kern-\nulldelimiterspace} {Q_{E} ^{n} } $$ over QE and QL is shown for most studies. Contrasting with an optimal n, another value n0 can also be found where D = 0. Occasionally, n0 can be close to 1, and RI then will have a small D and poor performance. Simulation with kinetic modeling of FDG dual-time scans for liver and liver metastases demonstrates worst and best scan times. Indicated for these imaging protocols are QE at very early cellular transport associated times and QL rather late when phosphorylation/dephosphorylation dominate. Benefits from choosing optimal times in dual-time protocols, especially in combination with choosing optimal ns, can be significant.ConclusionA protocol-dependent optimizing parameter n in an improved classification marker can easily be identified in a learning set of scans having normals and abnormals. Finding this parameter below 1.0 in most all studies suggests that a popularly used QL/QE may often overweight early activities. Additionally, QL/QE may sometimes be a poor marker choice and underestimate a protocol’s diagnostic capability. Subsequent use of the proposed $$ {Q_{L} } \mathord{\left/ {\vphantom {{Q_{L} } {Q_{E} ^{n} }}} \right. \kern-\nulldelimiterspace} {Q_{E} ^{n} } $$ in settings similar to that of the learning set gives improved diagnostic performance over traditional approaches, although by widely varying amounts. Additionally, a method of seeking optimal scan times is demonstrated and suggests significant gains in dual-time protocol performances are possible.