Automated subfield volumetry of hippocampus is desirable for use in temporal lobe epilepsy (TLE), but its utility has not been established. Automatic segmentation of hippocampal subfields (ASHS) and the new version of FreeSurfer software (ver.6.0) using high-resolution T2-weighted MR imaging are candidates for this volumetry. The aim of this study was to evaluate hippocampal subfields in TLE patients using ASHS as well as the old and new versions of FreeSurfer. We recruited 50 consecutive unilateral TLE patients including 25 with hippocampal sclerosis (TLE-HS) and 25 without obvious etiology (TLE-nonHS). All patients and 45 healthy controls underwent high-resolution T2-weighted and 3D-volume T1-weighted MRI scanning. We analyzed all of their MR images by FreeSurfer ver.5.3, ver.6.0 and ASHS. For each subfield, normalized z-scores were calculated and compared among groups. In TLE-HS groups, ASHS and FreeSurfer ver.6.0 revealed maximal z-scores in ipsilateral cornu ammonis (CA) 1, CA4 and dentate gyrus (DG), whereas in FreeSurfer ver.5.3 ipsilateral subiculum showed maximal z-scores. In TLE-nonHS group, there was no significant volume reduction by either ASHS or FreeSurfer. ASHS and the new version of FreeSurfer may have an advantage in compatibility with existing histopathological knowledge in TLE patients with HS compared to the old version of FreeSurfer (ver.5.3), although further investigations with pathological findings and/or surgical outcomes are desirable.

In addition to occipital hypoperfusion, preserved metabolism of the posterior cingulate gyri (PCG) relative to the precunei is known as the cingulate island sign (CIS) in the patients with dementia with Lewy bodies (DLB). CIS has been detected using [18F]fluorodeoxyglucose positron emission tomography but not using brain perfusion single-photon emission computed tomography (SPECT). The purpose of this study was to optimize brain perfusion SPECT to enable differentiation of DLB from Alzheimer’s disease (AD) using CIS and occipital hypoperfusion. Eighteen patients with probable DLB and 17 age-matched Pittsburgh compound B-positive patients with AD underwent technetium-99m ethyl cysteinate dimer SPECT. SPECT Z-score maps were generated using the easy Z-score imaging system (eZIS) analysis software (Matsuda H, Mizumura S, Nagao T, Ota T, Iizuka T, Nemoto K, Takemura N, Arai H, Homma A, AJNR Am J Neuroradiol 28(4):731–6, 2007), which included volumes of interest (VOIs) in which a group comparison between patients with AD and cognitively normal subjects revealed significant relative hypoperfusion. We used the Montreal Neurological Institute (MNI) space anatomical border to divide the bilateral PCG to precunei VOIs into two parts, the PCG and precunei. Z-scores in the PCG, precunei, and occipital areas and ratios were analysed and compared with receiver operating characteristic (ROC) curve analyses. The largest area under the curve (AUC) value for use in differentiating DLB from AD with the ratio of PCG to medial occipital was 0.87; the accuracy, sensitivity, and specificity were 85.7, 88.9, and 82.4 %, respectively. The AUC with the ratio of PCG to the precuneus was smaller, and it was 0.85, though no significant difference was observed between these two AUCs. The Z-score ratio of the PCG within the early-AD-specific VOI to medial-occipital area is clinically useful in discriminating demented patients with DLB from those with AD.

Psychosis is one of the most important psychiatric comorbidities in temporal lobe epilepsy (TLE), and its pathophysiology still remains unsolved. We aimed to explore the connectivity differences of structural neuroimaging between TLE with and without psychosis using a graph theoretical analysis, which is an emerging mathematical method to investigate network connections in the brain as a small-world system. We recruited 11 TLE patients with unilateral hippocampal sclerosis (HS) presenting psychosis or having a history of psychosis (TLE-P group). As controls, 15 TLE patients with unilateral HS without any history of psychotic episodes were also recruited (TLE-N group). For graph theoretical analysis, the normalized gray matter images of both groups were subjected to Graph Analysis Toolbox (GAT). As secondary analyses, each group was compared to 14 age- and gender-matched healthy subjects. The hub node locations were found predominantly in the ipsilateral hemisphere in the TLE-N group, and mainly on the contralateral side in the TLE-P group. The TLE-P group showed significantly higher characteristic path length, transitivity, lower global efficiency, and resilience to random or targeted attack than those of the TLE-N group. The regional comparison in betweenness centrality revealed significantly decreased connectivity in the contralateral temporal lobe, ipsilateral middle frontal gyrus, and bilateral postcentral gyri in the TLE-P group. The healthy subjects showed well-balanced nodes/edges distributions, similar metrics to TLE-N group except for higher small-worldness/modularity/assortativity, and various differences of regional betweenness/clustering. In TLE with psychosis, graph theoretical analysis of structural imaging revealed disrupted connectivity in the contralateral hemisphere. The network metrics suggested that the existence of psychosis can bring vulnerability and decreased efficiency of the whole-brain network. The sharp differences in structural networks between morphologically homogeneous groups are remarkable and may contribute to a better understanding of psychosis in TLE.

Background: Deep learning (DL)-based image quality improvement is a novel technique based on convolutional neural networks. The aim of this study was to compare the clinical value of ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG PET) images obtained with the DL method with those obtained using a Gaussian filter. Methods: Fifty patients with a mean age of 64.4 (range, 19–88) years who underwent ¹⁸F-FDG PET/CT between April 2019 and May 2019 were included in the study. PET images were obtained with the DL method in addition to conventional images reconstructed with three-dimensional time of flight-ordered subset expectation maximization and filtered with a Gaussian filter as a baseline for comparison. The reconstructed images were reviewed by two nuclear medicine physicians and scored from 1 (poor) to 5 (excellent) for tumor delineation, overall image quality, and image noise. For the semi-quantitative analysis, standardized uptake values in tumors and healthy tissues were compared between images obtained using the DL method and those obtained with a Gaussian filter. Results: Images acquired using the DL method scored significantly higher for tumor delineation, overall image quality, and image noise compared to baseline (P < 0.001). The Fleiss’ kappa value for overall inter-reader agreement was 0.78. The standardized uptake values in tumor obtained by DL were significantly higher than those acquired using a Gaussian filter (P < 0.001). Conclusions: Deep learning method improves the quality of PET images.