Improving the reliability of deep learning computational ghost imaging with prediction uncertainty based on neighborhood feature maps.

Defect inspection is required in various fields, and many researchers have attempted deep-learning algorithms for inspections. Deep-learning algorithms have advantages in terms of accuracy and measurement time; however, the reliability of deep-learning outputs is problematic in precision measurements. This study demonstrates that iterative estimation using neighboring feature maps can evaluate the uncertainty of the outputs and shows that unconfident error predictions have higher uncertainties. In ghost imaging using deep learning, the experimental results show that removing outputs with higher uncertainties improves the accuracy by approximately 15.7%.

Impact of time-of-flight on qualitative and quantitative analyses of myocardial perfusion PET studies using (13)N-ammonia.

BACKGROUND: The impact of time-of-flight (TOF) in myocardial perfusion (13)N-ammonia positron emission tomography (PET) is unclear. METHODS AND RESULTS: Twenty consecutive subjects underwent rest and adenosine stress (13)N-ammonia myocardial perfusion PET. Two sets of images were reconstructed using TOF-ordered subset expectation maximization (TOF-OSEM) and 3-dimensional row-action maximum likelihood algorithm (3D-RAMLA). Qualitative and quantitative analyses from the TOF-OSEM and 3D-RAMLA reconstructions were compared. Count profile curves revealed that TOF relatively increased the uptake of (13)N-ammonia at the lateral walls, and apical thinning was emphasized on the TOF images. Both segmental rest and stress myocardial blood flow (MBF) values were higher with TOF-OSEM use than with 3D-RAMLA use (rest MBF: 0.955 ± 0.201 vs 0.836 ± 0.185, P < .001; stress MBF: 2.149 ± 0.697 vs 2.058 ± 0.721, P < .001). The differentiation of MBF between reconstructions was more enhanced under rest conditions. Thus, segmental myocardial flow reserve (MFR) observed using TOF-OSEM reconstruction was lower than that observed using 3D-RAMLA (2.25 ± 0.57 vs 2.46 ± 0.75, P < .001). No remarkable differences were observed between segmental and territorial results. CONCLUSIONS: TOF increased lateral wall counts and emphasized apical thinning. Quantitatively, TOF reconstruction showed increased MBF, especially under relatively low perfusion conditions.