Abstract Purpose Multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) are two major demyelinating diseases affecting the central nervous system (CNS). The objective of this study is to evaluate the prevalence of pontine trigeminal nerve lesions in patients diagnosed with MS and NMOSD using MRI. Methods This retrospective study included patients diagnosed with MS or NMOSD between July 2018 and July 2023. MS patients were clinically diagnosed using the 2017 McDonald criteria, while NMOSD patients were those who met the 2015 International Panel for NMO Diagnosis (IPND) criteria and were positive for Aquaporin-4 Antibody (AQP4-Ab). Results The study included a total of 90 patients, with 45 diagnosed with MS and another 45 with NMOSD. Pontine trigeminal nerve lesions were observed in both MS and NMOSD, but were more prevalent in MS patients (20 % vs. 2 %, p = 0.008). Root entry zone (REZ) lesions were found in 4 of 45 MS patients, accounting for 9 % (95 % CI: 3 %–17 %), and were absent in the NMOSD group; however, there was no significant difference between the two groups (p = 0.12). Of the MS patients with pontine trigeminal nerve lesions, 6 out of 9 (63 %; 95 % CI, 36 %–98 %) exhibited bilateral lesions, which was significantly more prevalent compared to the NMOSD group (13 % vs. 0 %, p = 0.03). Conclusions The presence of pontine trigeminal nerve lesions, particularly when bilateral, are significantly more prevalent in MS patients than in those with NMOSD, suggesting their utility as a distinctive marker and potential diagnostic indicator specifically for MS.

Purpose: Our earlier investigation assessed the correlation between regional accumulations of a tau positron emission tomography (PET) tracer (THK5351) and cognitive dysfunction in the Alzheimer's disease (AD) continuum brain. We reported significant correlations between bilateral parietotemporal and left posterior cingulate/precuneus THK5351 accumulations and MiniMental State Examination (MMSE)/Alzheimer Disease Assessment Scale-Cognitive Subtest (ADAS) scores. Herein, we explored the relationship between these variables in non-AD pathologic brains. Methods: Thirty-one subjects with non-AD pathologic changes (mean age: 64.6±11.3 yrs) underwent three-dimensional MRI, fluoro-2-deoxyglucose (FDG)-(PET), Pittsburgh compound B (PiB)-amyloid PET, and THK5351-tau PET. All subjects completed the MMSE (mean±SD score: 27.0±4.1), and 20 subjects (mean age: 63.8±11.9 yrs) also completed the ADAS (9.7±9.3). A voxel-wise statistical analysis using statistical parametric mapping determined the correlations between each subject's FDG and THK imaging results and their MMSE and ADAS scores. Results: The voxel-wise statistical analysis revealed that FDG accumulations in the left temporal cortices were significantly positively correlated with the subjects' MMSE scores and significantly negatively correlated with their ADAS scores. A negative correlation was observed between the subjects' MMSE scores and THK5351 accumulations in the left temporal and medial frontal cortices and the right temporal pole. A positive correlation was observed between the subjects' ADAS scores and THK5351 accumulations in the left temporal cortices and the right anterior temporal lobe. Conclusions: Compared to FDG accumulation, THK5351 accumulations correlated more strongly with neuropsychological test scores in non-AD pathologic brains, likely due to THK5351's affinity for both tau and monoamine oxidase-B. THK5351 measurements may be valuable for clarifying the pathophysiology of non-AD pathologic changes in the brain.

Abstract We propose data augmentation using fractal images to train deep-learning models for medical image diagnosis. Deep learning models for image classification typically demand large datasets, obtaining which can be challenging in medical image diagnosis. Current approaches often involve pre-training of model parameters on natural image databases like ImageNet and fine-tuning of the parameters with specific medical image data. However, natural and medical images have distinct characteristics, which question the suitability of pre-training on natural image data. Moreover, the scalability of natural image databases is limited; thus, acquiring sufficient data for large-scale deep-learning modelsis difficult. In contrast, Kataoka et al. introduced a mathematical model for generating image data and demonstrated its effectiveness in pre-training for natural image classification.In this study, we employed a pre-trained model utilizing fractals among mathematical models and experimentally classified CT images of COVID-19 pneumonia. The experimental results demonstrated that this fractal-based pretraining model can achieve accuracy comparable to conventional natural image-based approaches. Fractal images are easily generated compared to the natural images. Furthermore, by adjusting the parameters, generating appropriate data for specific applications can be possible. This flexibility in generating data allows for customization and optimization of the model for different scenarios or specific requirements. We believe this approach holds promise in medical image diagnosis, where the number of samples is often limited.

Abstract Deep learning (DL) has recently attracted attention for data processing in positron emission tomography (PET). Attenuation correction (AC) without computed tomography (CT) data is one of the interests. Here, we present, to our knowledge, the first attempt to generate an attenuation map of the human head via Sim2Real DL-based tissue composition estimation from model training using only the simulated PET dataset. The DL model accepts a two-dimensional non-attenuation-corrected PET image as input and outputs a four-channel tissue-composition map of soft tissue, bone, cavity, and background. Then, an attenuation map is generated by a linear combination of the tissue composition maps and, finally, used as input for scatter+random estimation and as an initial estimate for attenuation map reconstruction by the maximum likelihood attenuation correction factor (MLACF), i.e., the DL estimate is refined by the MLACF. Preliminary results using clinical brain PET data showed that the proposed DL model tended to estimate anatomical details inaccurately, especially in the neck-side slices. However, it succeeded in estimating overall anatomical structures, and the PET quantitative accuracy with DL-based AC was comparable to that with CT-based AC. Thus, the proposed DL-based approach combined with the MLACF is also a promising CT-less AC approach.

Abstract Purpose Predicting progression of mild cognitive impairment (MCI) to Alzheimer’s disease (AD) or dementia with Lewy bodies (DLB) is important. We evaluated morphological and functional differences between MCI with Lewy bodies (MCI-LB) and MCI due to AD (MCI-AD), and a method for differentiating between these conditions using brain MRI and brain perfusion SPECT. Methods A continuous series of 101 subjects, who had visited our memory clinic and met the definition of MCI, were enrolled retrospectively. They were consisted of 60 MCI-LB and 41 MCI-AD subjects. Relative cerebral blood flow (rCBF) on SPECT images and relative brain atrophy on MRI images were evaluated. We performed voxel-based analysis and visually inspected brain perfusion SPECT images for regional brain atrophy, occipital hypoperfusion and the cingulate island sign (CIS), for differential diagnosis of MCI-LB and MCI-AD. Results MRI showed no significant differences in regional atrophy between the MCI-LB and MCI-AD groups. In MCI-LB subjects, occipital rCBF was significantly decreased compared with MCI-AD subjects (p < 0.01, family wise error [FWE]-corrected). Visual inspection of occipital hypoperfusion had sensitivity, specificity, and accuracy values of 100%, 73.2% and 89.1%, respectively, for differentiating MCI-LB and MCI-AD. Occipital hypoperfusion was offered higher diagnostic utility than the CIS. Conclusions The occipital lobe was the region with significantly decreased rCBF in MCI-LB compared with MCI-AD subjects. Occipital hypoperfusion on brain perfusion SPECT may be a more useful imaging biomarker than the CIS for visually differentiating MCI-LB and MCI-AD.

Abstract There are growing expectations for AI computer-aided diagnosis: computer-aided diagnosis (CAD) systems can be used to improve the accuracy of diagnostic imaging. However, it is not easy to collect large amounts of disease medical image data with lesion area annotations for the supervised learning of CAD systems. This study proposes an unsupervised local image feature extraction method running without such disease medical image datasets. Local image features are one of the key determinants of system performance. The proposed method requires only a normal image dataset that does not include lesions and can be collected easier than a disease dataset. The unsupervised features are extracted by applying multiple convolutional autoencoders to analyze various 2.5-dimensional images. The proposed method is evaluated by two kinds of problems: the detection of cerebral aneurysms in head MRA images and the detection of lung nodules in chest CT images. In both cases, the performance is high, showing an AUC of more than 0.96. These results show that the proposed method can automatically learn features that are useful for lesion recognition from lesion-free normal data, regardless of the type of image or lesion.

Introduction: Alzheimer’ disease (AD) is characterized by amyloid and tau deposition in prodromal stage and subsequent decrease of cerebral glucose metabolism as disease progresses. We previously proposed the voxel-wise metabolism to amyloid deposits ratio (MAR) image and evaluated its reliability for the diagnosis of AD. This time we produced the voxel-wise metabolism to tau deposits ratio (MTR) image and evaluated the utility of MTR image. Methods: Subjects with AD continuum, who underwent 18F-FDG-PET, 18F-flutemetamol amyloid PET and 18F-MK-6240 tau-PET for BATON (Blood-based Amyloid, Tau and Other Neuropathological Biomarkers Project) at our institution were included to this study. Consecutive 25 subjects with AD continuum (age: 73.7 ± 7.2, MMSE: 22.1 ± 3.7) who had positive amyloid deposition shown by amyloid PET were selected. After normalizing to a standard stereotactic space, the MTR image was created by dividing each FDG-PET image by corresponding tau-PET image using with voxel-wise interimage computation. We performed paired t test voxel wise comparison between individual FDG-PET image and MTR image. We also evaluated the correlations between the regional FDG standardized up take value ratio (SUVR), MK-6240 SUVR, and MTR value and the patients' MMSE scores. Results: In the voxel wise comparison, the MTR images showed more severely affected region in bilateral inferior temporal gyri (p < 0.001, FWE) than FDG image (Fig.1). The bilateral parietal and posterior cingulate/precuneus, and right frontal FDG SUVR and MMSE scores were positively correlated, and the bilateral occipital and central gyri MK6240 SUVR were negatively correlated with MMSE scores. There were significant positive correlations between bilateral frontal, parietal and left posterior cingulate/precuneus MTR value and MMSE scores. Conclusions: MTR image reflected not only cerebral tau deposition but the glucose hypometabolism and demonstrated that cerebral glucose metabolism decreases beyond the degree of tau accumulation in inferior temporal area of AD continuum brain. There are regional differences in glucose metabolism and tau deposition that correlate with cognitive dysfunction in AD continuum.

Background The aim of this study was to estimate the impact of respiratory and electrocardiogram (ECG)-gated FDG positron emission tomography (PET)/computed tomography (CT) on the diagnosis of cardiac sarcoidosis (CS). Methods and Results Imaging from thirty-one patients was acquired on a PET/CT scanner equipped with a respiratory- and ECG-gating system. Non-gated PET images and three kinds of gated PET/CT images were created from identical list-mode clinical PET data: respiratory-gated PET during expiration (EX), ECG-gated PET at end diastole (ED), and ECG-gated PET at end systole (ES). The maximum standardized uptake value (SUVmax) and cardiac metabolic volume (CMV) were measured, and the locations of FDG accumulation were analyzed using a polar map. The mean SUVmax of the subjects was significantly higher after application of either respiratory-gated or ECG-gated reconstruction. Conversely, the mean CMV was significantly lower following the application of respiratory-gated or ECG-gated reconstruction. The segment showing maximum accumulation was shifted to the adjacent segment in 25.8%, 38.7%, and 41.9% of cases in EX, ED, and ES images, respectively. Conclusion In FDG PET/CT scanning for the diagnosis of CS, gated scanning is likely to increase quantitative accuracy, but the effect depends on the location and synchronization method.

Abstract. Recently, the number of images for pre-training of deep learning models has been increasing, and large-scale data sets contain inappropriate images such as ethically inappropriate images, copyright infringement, and labeling errors. A method to solve these is by using a fractal database that generates images by mathematical formulas without using natural images. Our goal is to show that the classification accuracy obtained by pre-training with fractal images is comparable to natural images. In the experiments, we compare the performance on the tasks to classify CT images of COVID-19 pneumonia and regular pneumonia.

Objective: We evaluated the relationship between regional accumulations of the tau positron emission tomography (PET) tracer THK5351 and cognitive dysfunction in the Alzheimer's disease (AD) continuum. Methods: The cases of 18 patients with AD or mild cognitive impairment (MCI) due to AD who underwent three-dimensional MRI, fluoro-2-deoxyglucose (FDG)-(PET), Pittsburgh compound B (PiB)-amyloid PET, and THK5351-tau PET were analyzed. Their mean age was 70.6 ± 11.3, their mean Mini-Mental State Examination (MMSE) score was 22.3 ± 6.8, and their mean Alzheimer Disease Assessment Scale-Cognitive Subtest (ADAS) score was 12.5 ± 7.3. To determine the correlation between each patient's four imaging results and their MMSE and ADAS scores, we performed a voxel-wise statistical analysis with statistical parametric mapping (SPM). Results: The SPM analysis showed that the bilateral parietotemporal FDG accumulations and MMSE scores were positively correlated, and the bilateral parietotemporal FDG accumulations were negatively correlated with ADAS scores. There were significant correlations between bilateral parietotemporal and left posterior cingulate/precuneus THK5351 accumulations and MMSE/ADAS scores. Conclusion: In the AD brain, THK5351 correlates with neuropsychological test scores as well as or more additional than FDG due to its affinity for both tau and monoamine oxidase-B (MAO-B), and measurements of THK5351 may thus be useful in estimating the progression of AD.

Abstract Background This study evaluated the physical performance of a positron emission tomography (PET) system dedicated to the head and breast according to the National Electrical Manufacturers Association (NEMA) NU2-2012 standard. Methods The spatial resolution, sensitivity, scatter fraction, count rate characteristics, corrections for count losses and randoms, and image quality of the system were determined. All measurements were performed according to the NEMA NU2-2012 acquisition protocols, but image quality was assessed using a brain-sized phantom. Furthermore, scans of the three-dimensional (3D) Hoffmann brain phantom and mini-Derenzo phantom were acquired to allow visual evaluation of the imaging performance for small structures. Results The tangential, radial, and axial full width at half maximum (FWHM) at a 10-mm offset in half the axial field of view were measured as 2.3, 2.5, and 2.9 mm, respectively. The average system sensitivity at the center of the field of view and at a 10-cm radial offset was 7.18 and 8.65 cps/kBq, respectively. The peak noise-equivalent counting rate was 35.2 kcps at 4.8 kBq/ml. The corresponding scatter fraction at the peak noise-equivalent counting rate was 46.8%. The peak true rate and scatter fraction at 8.6 kBq/ml were 127.8 kcps and 54.3%, respectively. The percent contrast value for a 10-mm sphere was approximately 50%. On the 3D Hoffman brain phantom image, the structures of the thin layers composing the phantom were visualized on the sagittal and coronal images. On the mini-Derenzo phantom, each of the 1.6-mm rods was clearly visualized. Conclusion Taken together, these results indicate that the head- and breast-dedicated PET system has high resolution and is well suited for clinical PET imaging.

Purpose: Although some dementia with Lewy bodies (DLB) patients have Alzheimer disease (AD) pathology, there are no reports comparing AD and DLB by amyloid PET and tau PET in the same patient. The aim of this study was to investigate the regional difference of metabolic reduction, amyloid deposition and tau deposition between DLB and AD brain. Methods: Five consecutive subjects with DLB and five age and Mini-mental state examination score matched AD patients, who underwent both fluoro-2-deoxyglucose (FDG)-PET, Pittsburgh compound B (PiB)-amyloid PET and THK5351 (THK)-tau PET, were included in this study. Voxel-based statistical analysis was performed with Statistical Parametric Mapping for comparison with FDG, amyloid and tau accumulation between the DLB and AD brains. Results: FDG PET demonstrated that occipital metabolism in DLB group was significantly lower than that in AD group. PiB PET demonstrated that in the AD group the anterior to posterior cingulate cortices, supplementary motor area, and thalamic amyloid depositions were significantly higher than those in the DLB group. THK PET demonstrated that in the AD group the posterior cingulate cortices, posterior internal capsules, thalamic and peri-brainstem regions tau depositions were significantly higher than those in the DLB group. Conclusions: In the DLB brain, it is suggested that decreased regional glucose metabolism is unrelated to amyloid and tau depositions, and the amyloid and tau depositions in the cingulate gyri and thalamic regions in the AD brain are higher than those in DLB brain. Key words: dementia with Lewy bodies (DLB), glucose metabolism, amyloid, tau, positron emission tomography (PET).

Abstract We propose a method to detect primary and metastatic lesions with Fluorine−18 fluorodeoxyglucose (FDG) accumulation in the lung field, neck, mediastinum, and bony regions on the FDG-PET/CT images. To search for systemic lesions, various anatomical structures must be considered. The proposed method is addressed by using an extraction process for anatomical regions and a uniform lesion detection approach. The uniform approach does not utilize processes that reflect any region-specific anatomical aspects but has a machine-learnable framework. Therefore, it can work as a lesion detection process for a specific anatomical region if it machine-learns the specific region data. In this study, three lesion detection processes for the whole-body bone region, lung field, or neck-mediastinum region are obtained. These detection processes include lesion candidate detection and false positive (FP) candidate elimination. The lesion candidate detection is based on a voxel anomaly detection with a one-class support vector machine. The FP candidate elimination is performed using an AdaBoost classifier ensemble. The image features used by the ensemble are selected sequentially during training and are optimal for candidate classification. Three-fold cross-validation was used to detect performance with the 54 diseased FDG-PET/CT images. The mean sensitivity for detecting primary and metastatic lesions at 3 FPs per case was 0.89 with a 0.10 standard deviation (SD) in the bone region, 0.80 with a 0.10 SD in the lung field, and 0.87 with a 0.10 SD in the neck region. The average areas under the ROC curve were 0.887 with a 0.125 SD for detecting bone metastases, 0.900 with a 0.063 SD for detecting pulmonary lesions, and 0.927 with a 0.035 SD for detecting the neck-mediastinum lesions. These detection performances indicate that the proposed method could be applied clinically. These results also show that the uniform approach has high versatility for providing various lesion detection processes.

Objectives This study evaluates the phantom attenuation correction (PAC) method as an alternative to maximum-likelihood attenuation correction factor (ML-ACF) correction in time-of-flight (TOF) brain positron emission tomography (PET) studies. Methods In the PAC algorithm, a template emission image λRef and a template attenuation coefficient image μRef are prepared as a data set based on phantom geometry. Position-aligned attenuation coefficient image μAcq is derived by aligning μRef using parameters that match the template emission image λRef to measured emission image λAcq. Then, attenuation coefficient image μAcq combined with a headrest image is used for scatter and attenuation correction in the image reconstruction. To evaluate the PAC algorithm as an alternative to ML-ACF, Hoffman 3D brain and cylindrical phantoms were measured to obtain the image quality indexes of contrast and uniformity. These phantoms were also wrapped with a radioactive sheet to obtain attenuation coefficient images using ML-ACF. Emission images were reconstructed with attenuation correction by PAC and ML-ACF, and the results were compared using contrast and uniformity as well as visual assessment. CT attenuation correction (CT-AC) was also applied as a reference. Results The contrast obtained by ML-ACF was slightly overestimated due to its unique experimental condition for applying ML-ACF in Hoffman 3D brain phantom but the uniformity was almost equivalent among ML-ACF, CT-AC, and PAC. PAC showed reasonable result without overestimation compared to ML-ACF and CT-AC. Conclusions PAC is an attenuation correction method that can ensure the performance in phantom test, and is considered to be a reasonable alternative to clinically used ML-ACF-based attenuation correction.

Abstract We acquired brain positron emission tomography (PET) images of fluorodeoxyglucose (FDG) and flutemetamol PET using a time-of-flight-PET system dedicated for the head (dhPET) and a conventional whole-body PET/computed tomography (wbPET) system and evaluated the clinical superiority of dhPET over wbPET. Methods: There were 18 subjects for the FDG-PET study and 17 subjects for the flutemetamol PET study. FDG-PET images were first obtained using wbPET, followed by dhPET. Flutemetamol PET images were first obtained using wbPET, followed by dhPET. Images acquired using dhPET and wbPET were compared by visual inspection, voxel-wise analysis, and standard uptake value ratio (SUVR). Results: All FDG and flutemetamol images acquired using dhPET were judged as better by visual inspection than those acquired using wbPET. The voxel-wise analysis demonstrated that accumulations in the cerebellum, lateral occipital cortices, and around the central sulcus area in dhPET FDG images were lower than those in wbPET FDG images, whereas accumulations around the ventricle systems were higher in dhPET FDG images than those in wbPET FDG images. Accumulations in the cerebellar dentate nucleus, midbrain, lateral occipital cortices, and around the central sulcus area in dhPET images were lower than those in wbPET images, whereas accumulations around the ventricle systems were higher in dhPET flutemetamol images than those in wbPET flutemetamol images. Mean cortical SUVRs of FDG and flutemetamol dhPET images were significantly higher than those of FDG and flutemetamol wbPET images, respectively. Conclusion: The dhPET images had better image quality by visual inspection and higher SUVRs than wbPET images. Although there were several regional accumulation differences between dhPET and wbPET images, understanding this phenomenon will enable full use of the features of this dhPET system in clinical practice.

Abstract This study investigates the equivalence or compatibility between U-Net and visual segmentations of fibroglandular tissue regions by mammography experts for calculating the breast density and mean glandular dose (MGD). A total of 703 mediolateral oblique-view mammograms were used for segmentation. Two region types were set as the ground truth (determined visually) :(1) one type included only the region where fibroglandular tissue was identifiable (called the dense region); (2) the other type included the region where the fibroglandular tissue may have existed in the past, provided that apparent adipose-only parts, such as the retromammary space, are excluded (the diffuse region). U-Net was trained to segment the fibroglandular tissue region with an adaptive moment estimation optimiser, five-fold cross-validated with 400 training and 100 validation mammograms, and tested with 203 mammograms. The breast density and MGD were calculated using the van Engeland and Dance formulas, respectively, and compared between U-Net and the ground truth with the Dice similarity coefficient and Bland–Altman analysis. Dice similarity coefficients between U-Net and the ground truth were 0.895 and 0.939for the dense and diffuse regions, respectively. In the Bland–Altman analysis, no proportional or fixed errors were discovered in either the dense or diffuse region for breast density, whereas a slight proportional error was discovered in both regions for the MGD (the slopes of the regression lines were −0.0299 and−0.0443 for the dense and diffuse regions, respectively). Consequently, the U-Net and ground truth were deemed equivalent(interchangeable)for breast density and compatible (interchange-able following four simple arithmetic operations) for MGD. U-Net-based segmentation of the fibroglandular tissue region was satisfactory for both regions, providing reliable segmentation for breast density and MGD calculations. U-Net will be useful in developing a reliable individualised screening mammography programme, instead of relying on the visual judgement of mammography experts.

Abstract Objective: The objective of the current study was to develop a novel, artificial intelligence (AI)-based system to diagnose coronavirus disease (COVID-19) using computed tomography (CT) slice images. Prior research has demonstrated that, if not focused on the lungs, AI diagnoses COVID-19 using information outside the lungs. The inclusion of CT training data from multiple facilities and CT models may also cause AI to diagnose COVID-19 with features that are irrelevant to COVID-19. Thus, the objective of the current study was to evaluate a combination of lung mask images and CT slice images from a single facility, using a single CT model, and use AI to differentiate COVID-19 from other types of pneumonia based solely on information related to the lungs. Method: By superimposing lung mask images on image feature output using an existing AI structure, it was possible to exclude image features other than those around the lungs. The results of this model were also compared with the slice image findings from which only the lung region was extracted. The system adopted an ensemble approach. The outputs of multiple AIs were averaged to differentiate COVID-19 cases from other types of pneumonia, based on CT slice images. Results: The system evaluated 132 scans of COVID-19 cases and 62 scans of non-COVID-19 cases taken at the single facility using a single CT model. The initial sensitivity, specificity, and accuracy of our system, using a threshold value of 0.50, was shown to be 95%, 53%, and 81%, respectively. Setting the threshold value to 0.84 adjusted the sensitivity and specificity to clinically usable values of 76% and 84%, respectively. Conclusion: The system developed in the current study was able to differentiate between pneumonia due to COVID-19 and other types of pneumonia with sufficient accuracy for use in clinical practice. This was accomplished without the inclusion of images of clinically meaningless regions and despite the application of more stringent conditions, compared to prior studies.

Abstract Objective This study aimed at investigating the correlation between recurrent visual hallucinations (VHs) and regional cerebral blood flow (rCBF) in patients with dementia with Lewy bodies (DLB). Methods In 147 DLB patients, the correlation between noise pareidolia scores and rCBF in brain perfusion single photon emission computed tomography (SPECT) was evaluated. The 147 subjects comprised 52 probable and 95 possible DLB patients, of whom 107 did not have visual hallucinations and 40 had visual hallucinations. Brain perfusion SPECT was then performed, and memory impairment was assessed using the Mini-Mental State Examination (MMSE), while the optical illusion “pareidolia” (the tendency to see a specific image in a random visual pattern) was evaluated using noise pareidolia test. The correlations between rCBF and MMSE or noise pareidolia scores were then analyzed. Results The rCBF and MMSE scores were positively correlated, and rCBF was correlated with MMSE scores in a region that was consistent with a previously reported memory-related site. There was no correlation between noise pareidolia scores and occipital CBF, but there were weak correlations between noise pareidolia scores and rCBF in the bilateral frontal lobes (Brodmann area [BA]8 and BA9), left cingulate cortex (BA31), and left angular and supramarginal gyri (BA39 and BA40) in DLB patients. Conclusion Weak correlation was found between noise pareidolia scores and rCBF in several sites (BA8, BA9, BA31, BA39 and BA40) other than in occipital lobe. These findings suggest that DLB hallucinations may be manifested by more complex brain network disorders, rather than by primary visual cortex disorders alone.

Abstract Aim The aim of this study was to evaluate an image reconstruction algorithm, including a new maximum-likelihood attenuation correction factor (ML-ACF) for time of flight (TOF) brain positron emission tomography (PET). Methods The implemented algorithm combines an ML-ACF method that simultaneously estimates both the emission image and attenuation sinogram from TOF emission data, and a scaling method based on anatomical features. To evaluate the algorithm’s quantitative accuracy, three-dimensional brain phantom images were acquired and soft-tissue attenuation coefficients and emission values were analyzed. Results The heterogeneous distributions of attenuation coefficients in soft tissue, skull, and nasal cavity were sufficiently visualized. The attenuation coefficient of soft tissue remained within 5% of theoretical value. Attenuation-corrected emission showed no lateral differences, and significant differences among soft tissue were within the error range. Conclusion The ML-ACF-based attenuation correction implemented for TOF brain PET worked well and obtained practical levels of accuracy.

Abstract Purpose To determine the association between occipital amyloid-PET uptake and neurocognitive performance in Alzheimer’s disease (AD). Materials and methods Fifty-eight participants with normal aged, mild cognitive impairment (MCI) due to AD and AD subjects who underwent F-18 florbetapir brain PET/CT scans were divided into four groups (A, normal; B, MCI; C, mild AD; and D, moderate/severe AD). Semiquantitative analyses of SUVR images were performed. The differences between groups and the correlations between florbetapir uptake and Thai Mental State Examination (TMSE) scores were determined. Significant differences were defined using a P < 0.001, uncorrected, or a P < 0.05, FWE for the voxel-based analyses with Statistical Parametric Mapping (SPM). Results There was a slightly higher florbetapir uptake in the precuneus, parietal, and occipital association cortices in Group B > A. The occipital florbetapir uptake in Groups C and D was significantly higher than in Group A, in addition to the precuneus, anterior cingulate, posterior cingulate, temporoparietal, and frontal cortices. There was a strong negative correlation between TMSE scores and florbetapir uptake in the occipital lobe. Conclusions Occipital amyloid uptake is associated with clinically advanced AD, and is inversely correlated with neurocognitive performance and may be useful for evaluating AD severity.

Abstract Logan graphical analysis (LGA) is a method for in vivo quantification of tracer kinetics in positron emission tomography (PET). The shortcoming of LGA is the presence of a negative bias in the estimated parameters for noisy data. Various approaches have been proposed to address this issue. We recently applied an alternative regression method called least-squares cubic (LSC), which considers the errors in both the predictor and response variables to estimate the LGA slope. LSC reduced the bias in non-displaceable binding potential estimates while causing slight increases in the variance. In this study, we combined LSC with a principal component analysis (PCA) denoising technique to counteract the effects of variance on parametric image quality, which was assessed in terms of the contrast between gray and white matter. Tissue time–activity curves were denoised through PCA, prior to estimating the regression parameters using LSC. We refer to this approach as LSC–PCA. LSC–PCA was assessed against OLS–PCA (PCA with ordinary least-squares (OLS)), LSC, and conventional OLS-based LGA. Comparisons were made for simulated 11 C-carfentanil and 11C Pittsburgh compound B (11C-PiB) data, and clinical 11 C-PiB PET images. PCA-based methods were compared over a range of principal components, varied by the percentage variance they account for in the data. The results showed reduced variances in distribution volume ratio estimates in the simulations for LSC–PCA compared to LSC, and lower bias compared to OLS–PCA and OLS. Contrasts were not significantly improved in clinical data, but they showed a significant improvement in simulation data —indicating a potential advantage of LSC–PCA over OLS–PCA. The effects of bias reintroduction when many principal components are used were also observed in OLS–PCA clinical images. We therefore encourage the use of LSC–PCA. LSC–PCA can allow the use of many principal components with minimal risk of bias, thereby strengthening the interpretation of PET parametric images.

Purpose It is usually easy to judge whether amyloid PET images should be interpreted as positive or negative for amyloid deposits by visual inspection or quantitative measurement standard uptake value ratio (SUVR), but the findings are equivocal in some cases. As conventional mean cortical SUVR (mcSUVR) measures accumulation in both gray matter (GM) and white matter, it may mis-estimate amyloid deposits. The purpose of the study was to develop a regional GM-dedicated SUVR measuring (GMSUVR) system for amyloid PET images with 3D-MRI, and evaluate its utility for detecting amyloid deposits in equivocal cases. Methods Of 126 subjects who underwent amyloid PET with 11C-PiB and 3D-MRI, the area of amyloid-positive regions and the critical regional GMSUVR thresholds were first determined in 15 amyloid-positive and 15 amyloid-negative patients, using the automatic volumetric measurement of segmented brain images system. We then tested 36 amyloid-negative, 60 amyloid-positive, and 13 equivocal subjects with this GMSUVR system and with conventional mcSUVR. Results Sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV) were 100%, 92%, 97%, 95%, and 100% for the GMSUVR system; and 97%, 86%, 93%, 92% and 94%, respectively, for mcSUVR. In 24 cases in which the findings were equivocal or discordant, the sensitivity, specificity, accuracy, PPV, and NPV were all 100% for the GMSUVR system; and were 90%, 33%, 83%, 90%, and 33%, respectively, for mcSUVR. Conclusion The regional GMSUVR measurement method was well able to discriminate between amyloid-positive and -negative subjects, even in cases where amyloid deposition was equivocal.

Abstract Objective An artificial intelligence (AI)-based algorithm typically requires a considerable amount of training data; however, few training images are available for dementia with Lewy bodies and frontotemporal lobar degeneration. Therefore, this study aims to present the potential of cycle-consistent generative adversarial networks (CycleGAN) to obtain enough number of training images for AI-based computer-aided diagnosis (CAD) algorithms for diagnosing dementia. Methods We trained CycleGAN using 43 amyloid-negative and 45 positive images in slice-by-slice. Results The CycleGAN can be used to synthesize reasonable amyloid-positive images, and the continuity of slices was preserved. Discussion Our results show that CycleGAN has the potential to generate a sufficient number of training images for CAD of dementia.

Background The Logan graphical analysis (LGA) algorithm is widely used to quantify receptor density for parametric imaging in positron emission tomography (PET). Estimating receptor density, in terms of the non-displaceable binding potential (BPND), from the LGA using the ordinary least-squares (OLS) method has been found to be negatively biased owing to noise in PET data. This is because OLS does not consider errors in the X-variable (predictor variable). Existing bias reduction methods can either only reduce the bias slightly or reduce the bias accompanied by increased variation in the estimates. In this study, we addressed the bias reduction problem by applying a different regression method. Methods We employed least-squares cubic (LSC) linear regression, which accounts for errors in both variables as well as the correlation of these errors. Noise-free PET data were simulated, for 11C-carfentanil kinetics, with known BPND values. Statistical noise was added to these data and the BPNDs were re-estimated from the noisy data by three methods, conventional LGA, multilinear reference tissue model 2 (MRTM2), and LSC-based LGA; the results were compared. The three methods were also compared in terms of beta amyloid (A β) quantification of 11C-Pittsburgh compound B brain PET data for two patients with Alzheimer’s disease and differing A β depositions. Results Amongst the three methods, for both synthetic and actual data, LSC was the least biased, followed by MRTM2, and then the conventional LGA, which was the most biased. Variations in the LSC estimates were smaller than those in the MRTM2 estimates. LSC also required a shorter computational time than MRTM2. Conclusions The results suggest that LSC provides a better trade-off between the bias and variability than the other two methods. In particular, LSC performed better than MRTM2 in all aspects; bias, variability, and computational time. This makes LSC a promising method for BPND parametric imaging in PET studies.

OBJECTIVE: This study aims to develop an algorithm named AutoRef to delineate a reference region for quantitative PET amyloid imaging. METHODS: AutoRef sets the reference region automatically using a distinguishing feature in the kinetics of reference region. This is reflected in the shapes of the tissue time activity curve. A statistical shape recognition algorithm of the gaussian mixture model is applied with considering spatial and temporal information on a reference region. We evaluate the BPND with manually set reference region and AutoRef using 86 cases (43 positive cases, 10 equivocal cases, and 33 negative cases) of dynamically scanned 11C-Pittsburgh Compound-B. RESULTS: From the Bland-Altman plot, the difference between two BPND is 0.099 ± 0.21 as standard deviation, and no significant systematic error is observed between the BPND with AutoRef and with manual definition of a reference region. Although a proportional error is detected, it is smaller than the 95% limits of agreement. Therefore, the proportional error is negligibly small. CONCLUSIONS: AutoRef presents the same performance as the manual definition of the reference region. Further, since AutoRef is more algorithmic than the ordinary manual definition of the reference region, there are few operator-oriented uncertainties in AutoRef. We thus conclude that AutoRef can be applied as an automatic delineating algorithm for the reference region in amyloid imaging.

画像診断の多くは医師による目視で行われているが，病変部の見落としによる治療の遅れが問題となっている．そこで，医師の負担を減らし病気の早期発見，診断の補助を目的とし，機械学習を用いた胸部CT画像における肺炎識別を行う．機械学習で十分な精度を得るためには，十分な数の事前学習用データセットが必要になる．本研究では，自然画像を用いず数式により事前学習用画像を生成するフラクタルデータベースを用いて従来の自然画像を用いた事前学習と遜色ない結果が得られることを示す．

Abstract. Recently, the number of images for pre-training of deep learning models has been increasing, and large-scale data sets contain inappropriate images such as ethically inappropriate images, copyright infringement, and labeling errors. A method to solve these is by using a fractal database that generates images by mathematical formulas without using natural images. Our goal is to show that the classification accuracy obtained by pre-training with fractal images is comparable to natural images. In the experiments, we compare the performance on the tasks to classify CT images of COVID-19 pneumonia and regular pneumonia.

Many studies have assessed breast density in clinical practice. However, calculation of breast density requires segmentation of the mammary gland region, and deep learning has only recently been applied. Thus, the robustness of the deep learning model for different image processing types has not yet been reported. We investigated the accuracy of segmentation of the U-net for mammograms made with variousimage processing types. We used 478 mediolateral oblique view mammograms. The mammograms were divided into 390 training images and 88 testing images. The ground truth of the mammary gland region made by mammary experts was used for the training and testing datasets. Four types of image processing (Types 1–4) were applied to the testing images to compare breast density in the segmented mammary gland regions with that of ground truths. The shape agreement between ground truth and the segmented mammary gland region by U-net of Types 1–4 was assessed using the Dice coefficient, and the equivalence or compatibility of breast density with ground truth was assessed by Bland-Altman analysis. The mean Dice coefficients between the ground truth and U-net were 0.952, 0.948, 0.948, and 0.947 for Types 1, 2, 3, and 4, respectively. By Bland-Altman analysis, the equivalence of breast density between ground truth and U-net was confirmed for Types 1 and 2, and compatibility was confirmed for Types 3 and 4. We concluded that the robustness of the U-net for segmenting the mammary gland region was confirmed for different image processing types.