TY - JOUR T1 - <strong>Validation of a Source-Less, Deep Learning Technique for Attenuation Correction of Stress-only CZT-SPECT studies.</strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3386 LP - 3386 VL - 63 IS - supplement 2 AU - James Case AU - Friedman John AU - A. McGhie AU - Prem Soman AU - Staci Courter AU - Katrina Case AU - Ronald Schwartz AU - Daniel Berman AU - Timothy Bateman Y1 - 2022/06/01 UR - http://jnm.snmjournals.org/content/63/supplement_2/3386.abstract N2 - 3386 Introduction: The implementation of highly efficient and accurate stress-only myocardial perfusion SPECT protocols has been challenging because of the limited availability of attenuation correction (AC) on SPECT devices. This study examined the clinical utility of a source-less AC (SLAC) technique for correcting attenuation artifact. The primary study end-point was the relative % of stress-only studies needing a follow-up rest scan, non-AC (NONAC) vs SLAC vs CTAC.Methods: 100 patients (75 male, mean BMI 29±4 with range 21-39 kg/m2, age=64±10) were included in this study. Patients received regadenoson stress followed by upright and supine imaging on a Spectrum Dynamics CZT system. Each patient also received a low-dose CT acquired on a separate system for AC. NONAC images sets were created using the system’s proprietary iterative reconstruction. For CTAC, CT maps were registered to the supine emission data and then reconstructed. For SLAC, supine emissions images were attenuation corrected using a proprietary deep learning algorithm. The following image sets were evaluated by a blinded expert panel for the need for rest imaging, overall diagnosis (Normal, Abnormal and Equivocal), and each of 17 segments (normal vs abnormal): NONAC supine plus upright; CTAC supine plus upright; SLAC supine plus upright. . Reader results were combined by using majority consensus. Results: Agreement on perceived need for a rest study was highest for SLAC (75%) compared to 65% for CTAC and 54% for NONAC, , though these differences did not achieve statistical significance (minimum p value, p(SLAC vs NONAC) = 0.06). Overall diagnosis was the same for CTAC and SLAC studies in 98% of the studies. All three readers agreed on the global diagnosis in 85% of CTAC and SLAC studies, vs 73% for NONAC (p=0.34). Concordance between both AC methods was high and statistically higher vs NONAC: NONAC vs CTAC studies (91/97, p=0.004) and NONAC vs SLAC studies (90/97, p=0.0003). A summary of the rates of perceived abnormal findings and need for a rest scan are given in Table 1.Segmentally, the SLAC and CTAC assessment of normal and abnormal agreed 86% of the time while SLAC vs NONAC agreement was 59% (p&lt;10-14), and and 55% for CTAC and NONAC (p&lt;10-19).Conclusions: CONCLUSION: Attenuation correction of the stress-first images, whether performed with CT or with this emission-based technique, resulted in a similar and lower rate of perceived need for a rest scan. CTAC and SLAC performed similarly, suggesting that this more efficient software-based method of attenuation correction could be mass-marketed to enhance adoption of stress-first and stress-only if normal imaging. ER -