Diagnostic approach with Z-score mapping to reduce artifacts caused by cerebral atrophy in regional CBF assessment of mild cognitive impairment (MCI) and Alzheimer's disease by [99mTc]-ECD and SPECT.

PURPOSE: The aim of this study was to develop a novel approach that enhanced diagnostic accuracy in the diagnosis of mild cognitive impairment (MCI) and early Alzheimer's disease (AD) using cerebral perfusion SPECT by minimizing artifacts caused by cerebral atrophy. MATERIALS AND METHODS: [99mTc]-ECD and SPECT studies were performed on 15 cognitively normal patients, 40 patients with MCI, and 16 patients with AD. SPECT images were compared using SPM. The atrophy correction method was incorporated to reduce artifacts through the MRI masking procedure. Regional Z-score, percent extent, and atrophy correction rate were obtained and compared. The Z-score mapping program was structured as a single package that ran semi-automatically. RESULTS: The method significantly reduced regional Z-score in most regions, leading to improved estimates. The mean atrophy correction rate ranged from 10.4 to 12.0%. In MCI and AD, the convexities of the frontal and parietal lobes and the posterior medial cerebrum were particularly sensitive to cerebral atrophy, and the Z-scores were overestimated, whereas the posterior cingulate cortex and the cerebellum were less sensitive. The diagnostic accuracy for MCI increased from 67 to 69% and for AD from 78 to 82%. CONCLUSION: The proposed approach provided more precise Z-scores with less over- or underestimation, artifacts, and improved diagnostic accuracy, being recommended for clinical studies.

Machine learning-based prediction of conversion coefficients for I-123 metaiodobenzylguanidine heart-to-mediastinum ratio.

PURPOSE: We developed a method of standardizing the heart-to-mediastinal ratio in 123I-labeled meta-iodobenzylguanidine (MIBG) images using a conversion coefficient derived from a dedicated phantom. This study aimed to create a machine-learning (ML) model to estimate conversion coefficients without using a phantom. METHODS: 210 Monte Carlo (MC) simulations of 123I-MIBG images to obtain conversion coefficients using collimators that differed in terms of hole diameter, septal thickness, and length. Simulated conversion coefficients and collimator parameters were prepared as training datasets, then a gradient-boosting ML was trained to estimate conversion coefficients from collimator parameters. Conversion coefficients derived by ML were compared with those that were MC simulated and experimentally derived from 613 phantom images. RESULTS: Conversion coefficients were superior when estimated by ML compared with the classical multiple linear regression model (root mean square deviations: 0.021 and 0.059, respectively). The experimental, MC simulated, and ML-estimated conversion coefficients agreed, being, respectively, 0.54, 0.55, and 0.55 for the low-; 0.74, 0.70, and 0.72 for the low-middle; and 0.88, 0.88, and 0.88 for the medium-energy collimators. CONCLUSIONS: The ML model estimated conversion coefficients without the need for phantom experiments. This means that conversion coefficients were comparable when estimated based on collimator parameters and on experiments.

Automatic quantification package (Hone Graph) for phantom-based image quality assessment in bone SPECT: computerized automatic classification of detectability.

OBJECTIVE: We previously developed a custom-design thoracic bone scintigraphy-specific phantom ("SIM2 bone phantom") to assess image quality in bone single-photon emission computed tomography (SPECT). We aimed to develop an automatic assessment system for imaging technology in bone SPECT and demonstrate the validity of this system. METHODS: Four spherical lesions of 13-, 17-, 22-, and 28-mm diameters in the vertebrae of SIM2 bone phantom simulating the thorax were filled with radioactivity (target-to-background ratio: 4). Dynamic SPECT acquisitions were performed for 15 min; reconstructions were performed using ordered subset expectation maximization at 3-15-min timepoints. Consequently, 216 lesions (54 SPECT images) were obtained: 120 and 96 lesions were used for software development and validation, respectively. The developed software used statistical parametric mapping to rigidly register and automatically calculate quantitative indexes (contrast-to-noise ratio, % coefficient of variance, % detectability equivalence volume, recovery coefficient, target-to-normal bone ratio, and full width at half maximum). A detectability score (DS) was used to define the four observation types (4, excellent; 3, adequate; 2, average; 1, poor) to score hot spherical lesions. The gold standard for DSs was independently classified by three experienced board-certified nuclear medicine technologists using the four observation types; thereafter, a consensus regarding the gold standard for DSs was reached. Using 120 lesions for development, decision tree analysis was performed to determine DS based on the quantitative indexes. We verified the validation of the quantitative indexes and their threshold values for automatic classification using 96 lesions for validation. RESULTS: The trends in the automatically calculated quantitative indices were consistent. Decision tree analysis produced four terminal groups; two quantitative indexes (% detectability equivalence volume and contrast-to-noise ratio) were used to classify DS. The automatically classified DSs exhibited an almost perfect agreement with the gold standard. The percentage agreement and kappa coefficient were 91.7% and 0.93, respectively, in 96 lesions for validation. CONCLUSIONS: The developed software automatically classified the detectability of hot lesions in the SIM2 bone phantom using the automatically calculated quantitative indexes, suggesting that this software could provide a means to automatically perform detectability analysis after data input that is excellent in reproducibility and accuracy.