Deep Learning-based Hierarchical Brain Segmentation with Preliminary Analysis of the Repeatability and Reproducibility.

PURPOSE: We developed new deep learning-based hierarchical brain segmentation (DLHBS) method that can segment T1-weighted MR images (T1WI) into 107 brain subregions and calculate the volume of each subregion. This study aimed to evaluate the repeatability and reproducibility of volume estimation using DLHBS and compare them with those of representative brain segmentation tools such as statistical parametric mapping (SPM) and FreeSurfer (FS). METHODS: Hierarchical segmentation using multiple deep learning models was employed to segment brain subregions within a clinically feasible processing time. The T1WI and brain mask pairs in 486 subjects were used as training data for training of the deep learning segmentation models. Training data were generated using a multi-atlas registration-based method. The high quality of training data was confirmed through visual evaluation and manual correction by neuroradiologists. The brain 3D-T1WI scan-rescan data of the 11 healthy subjects were obtained using three MRI scanners for evaluating the repeatability and reproducibility. The volumes of the eight ROIs-including gray matter, white matter, cerebrospinal fluid, hippocampus, orbital gyrus, cerebellum posterior lobe, putamen, and thalamus-obtained using DLHBS, SPM 12 with default settings, and FS with the "recon-all" pipeline. These volumes were then used for evaluation of repeatability and reproducibility. RESULTS: In the volume measurements, the bilateral thalamus showed higher repeatability with DLHBS compared with SPM. Furthermore, DLHBS demonstrated higher repeatability than FS in across all eight ROIs. Additionally, higher reproducibility was observed with DLHBS in both hemispheres of six ROIs when compared with SPM and in five ROIs compared with FS. The lower repeatability and reproducibility in DLHBS were not observed in any comparisons. CONCLUSION: Our results showed that the best performance in both repeatability and reproducibility was found in DLHBS compared with SPM and FS.

Reducing image artifacts in sparse projection CT using conditional generative adversarial networks.

Reducing the amount of projection data in computed tomography (CT), specifically sparse-view CT, can reduce exposure dose; however, image artifacts can occur. We quantitatively evaluated the effects of conditional generative adversarial networks (CGAN) on image quality restoration for sparse-view CT using simulated sparse projection images and compared them with autoencoder (AE) and U-Net models. The AE, U-Net, and CGAN models were trained using pairs of artifacts and original images; 90% of patient cases were used for training and the remaining for evaluation. Restoration of CT values was evaluated using mean error (ME) and mean absolute error (MAE). The image quality was evaluated using structural image similarity (SSIM) and peak signal-to-noise ratio (PSNR). Image quality improved in all sparse projection data; however, slight deformation in tumor and spine regions was observed, with a dispersed projection of over 5°. Some hallucination regions were observed in the CGAN results. Image resolution decreased, and blurring occurred in AE and U-Net; therefore, large deviations in ME and MAE were observed in lung and air regions, and the SSIM and PSNR results were degraded. The CGAN model achieved accurate CT value restoration and improved SSIM and PSNR compared to AE and U-Net models.

An analysis of the effectiveness of reflective learning through watching videos recorded with smart glasses-With multiple views (student, patient, and overall) in radiography education.

The Objective Structured Clinical Examination (OSCE) is designed to assess medical students' skills and attitude competencies before clinical practice. However, no method of reflective learning using video-based content has been used in OSCE education. This study aimed to confirm whether using smart glasses-based educational content is effective for OSCE reflective learning using multiple views (patient, student, and overall). This educational intervention study included a control group exposed to the traditional learning method and an intervention group exposed to a learning method incorporating smart glasses. Participants were 117 (72 in the control group and 45 in the intervention group) third-year radiological technology students scheduled to take the OSCE and 70 (37 in the control group and 33 in the intervention group) who met the eligibility criteria. Mock OSCEs were administered before and after the educational intervention (traditional and smart glasses-based education) to investigate changes in scores. After the educational intervention, a self-reported comprehension survey and a questionnaire were administered on the effectiveness of the video-based content from different views for student reflective learning. Unexpectedly, the OSCE evaluation score after the preliminary investigation significantly increased for the smart glasses control group (0.36±0.1) compared to the intervention group (0.06±0.1) setting up the radiographic conditions (x-ray center and detector center; p = 0.042). The intervention group's lower score in the mock OSCEs may have been due to the discomfort of wearing the smart glasses to perform the radiography procedure and their unfamiliarity with the smart glasses, which may have affected their concentration. The findings suggest that smart glasses-based education for OSCEs can be improved (e.g., being easy to handle and use and trouble-free).

Cross-sectional study on public health knowledge among first-year university students in Japan: Implications for educators and educational institutions.

In recent years, there have been increasing knowledge gaps and biases in public health information. This has become especially evident during the COVID-19 pandemic and has contributed to the spread of misinformation. With constant exposure to disinformation and misinformation through television, the internet, and social media, even university students studying healthcare-related subjects lack accurate public health knowledge. This study aimed to assess university students' knowledge levels of basic public health topics before they started their specialized education. Participants in this cross-sectional study were first-year students from medical schools, health-related colleges, and liberal arts colleges. A self-administered electronic survey was conducted from April to May 2021 at a private university in Japan, comprising six colleges with seven programs. Data analysis, conducted from June to December 2022, included students' self-reported public health knowledge, sources of information, and self-assessment of knowledge levels. Among the 1,562 students who received the questionnaire, 549 (192 male [35%], 353 female [64.3%], and 4 undisclosed [0.7%]) responded to one question (participants' response rate for each question; 59.6%-100%). The results showed that students had limited public health knowledge, especially in sexual health topics, and 10% of students reported not learning in class before university admission the following 11 topics: two on Alcohol, Tobacco, and Other Drugs; eight on Growth, Development, and Sexual Health; and one on Personal and Community Health. These results indicate significant knowledge gaps and biases, as well as gender gaps, in public health education, especially in the area of sexual health, which may help educators and educational institutions to better understand and prepare for further specialized education. The findings also suggest a need to supplement and reinforce the foundation of public health knowledge for healthcare majors at the time of university admission.

Accuracy of skull stripping in a single-contrast convolutional neural network model using eight-contrast magnetic resonance images.

In automated analyses of brain morphometry, skull stripping or brain extraction is a critical first step because it provides accurate spatial registration and signal-intensity normalization. Therefore, it is imperative to develop an ideal skull-stripping method in the field of brain image analysis. Previous reports have shown that convolutional neural network (CNN) method is better at skull stripping than non-CNN methods. We aimed to evaluate the accuracy of skull stripping in a single-contrast CNN model using eight-contrast magnetic resonance (MR) images. A total of 12 healthy participants and 12 patients with a clinical diagnosis of unilateral Sturge-Weber syndrome were included in our study. A 3-T MR imaging system and QRAPMASTER were used for data acquisition. We obtained eight-contrast images produced by post-processing T1, T2, and proton density (PD) maps. To evaluate the accuracy of skull stripping in our CNN method, gold-standard intracranial volume (ICVG) masks were used to train the CNN model. The ICVG masks were defined by experts using manual tracing. The accuracy of the intracranial volume obtained from the single-contrast CNN model (ICVE) was evaluated using the Dice similarity coefficient [= 2(ICVE ⋂ ICVG)/(ICVE + ICVG)]. Our study showed significantly higher accuracy in the PD-weighted image (WI), phase-sensitive inversion recovery (PSIR), and PD-short tau inversion recovery (STIR) compared to the other three contrast images (T1-WI, T2-fluid-attenuated inversion recovery [FLAIR], and T1-FLAIR). In conclusion, PD-WI, PSIR, and PD-STIR should be used instead of T1-WI for skull stripping in the CNN models.

Evaluation of motion artefact reduction depending on the artefacts' directions in head MRI using conditional generative adversarial networks.

Motion artefacts caused by the patient's body movements affect magnetic resonance imaging (MRI) accuracy. This study aimed to compare and evaluate the accuracy of motion artefacts correction using a conditional generative adversarial network (CGAN) with an autoencoder and U-net models. The training dataset consisted of motion artefacts generated through simulations. Motion artefacts occur in the phase encoding direction, which is set to either the horizontal or vertical direction of the image. To create T2-weighted axial images with simulated motion artefacts, 5500 head images were used in each direction. Of these data, 90% were used for training, while the remainder were used for the evaluation of image quality. Moreover, the validation data used in the model training consisted of 10% of the training dataset. The training data were divided into horizontal and vertical directions of motion artefact appearance, and the effect of combining this data with the training dataset was verified. The resulting corrected images were evaluated using structural image similarity (SSIM) and peak signal-to-noise ratio (PSNR), and the metrics were compared with the images without motion artefacts. The best improvements in the SSIM and PSNR were observed in the consistent condition in the direction of the occurrence of motion artefacts in the training and evaluation datasets. However, SSIM > 0.9 and PSNR > 29 dB were accomplished for the learning model with both image directions. The latter model exhibited the highest robustness for actual patient motion in head MRI images. Moreover, the image quality of the corrected image with the CGAN was the closest to that of the original image, while the improvement rates for SSIM and PSNR were approximately 26% and 7.7%, respectively. The CGAN model demonstrated a high image reproducibility, and the most significant model was the consistent condition of the learning model and the direction of the appearance of motion artefacts.

Analysis of synthetic magnetic resonance images by multi-channel segmentation increases accuracy of volumetry in the putamen and decreases mis-segmentation in the dural sinuses.

BACKGROUND: Voxel-based morphometry (VBM) using magnetic resonance imaging (MR) has been used to estimate cortical atrophy associated with various diseases. However, there are mis-segmentations of segmented gray matter image in VBM. PURPOSE: To study a twofold evaluation of single- and multi-channel segmentation using synthetic MR images: (1) mis-segmentation of segmented gray matter images in transverse and cavernous sinuses; and (2) accuracy and repeatability of segmented gray matter images. MATERIAL AND METHODS: A total of 13 healthy individuals were scanned with 3D quantification using an interleaved Look-Locker acquisition sequence with a T2 preparation pulse (3D-QALAS) sequence on a 1.5-T scanner. Three of the 13 healthy participants were scanned five consecutive times for evaluation of repeatability. We used SyMRI software to create images with three contrasts: T1-weighted (T1W), T2-weighted (T2W), and proton density-weighted (PDW) images. Manual regions of interest (ROI) on T1W imaging were individually set as the gold standard in the transverse sinus, cavernous sinus, and putamen. Single-channel (T1W) and multi-channel (T1W + T2W, T1W + PDW, and T1W + T2W + PDW imaging) segmentations were performed with statistical parametric mapping 12 software. RESULTS: We found that mis-segmentations in both the transverse and cavernous sinuses were large in single-channel segmentation compared with multi-channel segmentations. Furthermore, the accuracy of segmented gray matter images in the putamen was high in both multi-channel T1W + PDW and T1W + T2W + PDW segmentations compared with other segmentations. Finally, the highest repeatability of left putamen volumetry was found with multi-channel segmentation T1WI + PDWI. CONCLUSION: Multi-channel segmentation with T1WI + PDWI provides good results for VBM compared with single-channel and other multi-channel segmentations.

A cycle generative adversarial network for improving the quality of four-dimensional cone-beam computed tomography images.

BACKGROUND: Four-dimensional cone-beam computed tomography (4D-CBCT) can visualize moving tumors, thus adaptive radiation therapy (ART) could be improved if 4D-CBCT were used. However, 4D-CBCT images suffer from severe imaging artifacts. The aim of this study is to investigate the use of synthetic 4D-CBCT (sCT) images created by a cycle generative adversarial network (CycleGAN) for ART for lung cancer. METHODS: Unpaired thoracic 4D-CBCT images and four-dimensional multislice computed tomography (4D-MSCT) images of 20 lung-cancer patients were used for training. High-quality sCT lung images generated by the CycleGAN model were tested on another 10 cases. The mean and mean absolute errors were calculated to assess changes in the computed tomography number. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used to compare the sCT and original 4D-CBCT images. Moreover, a volumetric modulation arc therapy plan with a dose of 48 Gy in four fractions was recalculated using the sCT images and compared with ideal dose distributions observed in 4D-MSCT images. RESULTS: The generated sCT images had fewer artifacts, and lung tumor regions were clearly observed in the sCT images. The mean and mean absolute errors were near 0 Hounsfield units in all organ regions. The SSIM and PSNR results were significantly improved in the sCT images by approximately 51% and 18%, respectively. Moreover, the results of gamma analysis were significantly improved; the pass rate reached over 90% in the doses recalculated using the sCT images. Moreover, each organ dose index of the sCT images agreed well with those of the 4D-MSCT images and were within approximately 5%. CONCLUSIONS: The proposed CycleGAN enhances the quality of 4D-CBCT images, making them comparable to 4D-MSCT images. Thus, clinical implementation of sCT-based ART for lung cancer is feasible.

Advantages of Using Both Voxel- and Surface-based Morphometry in Cortical Morphology Analysis: A Review of Various Applications.

Surface-based morphometry (SBM) is extremely useful for estimating the indices of cortical morphology, such as volume, thickness, area, and gyrification, whereas voxel-based morphometry (VBM) is a typical method of gray matter (GM) volumetry that includes cortex measurement. In cases where SBM is used to estimate cortical morphology, it remains controversial as to whether VBM should be used in addition to estimate GM volume. Therefore, this review has two main goals. First, we summarize the differences between the two methods regarding preprocessing, statistical analysis, and reliability. Second, we review studies that estimate cortical morphological changes using VBM and/or SBM and discuss whether using VBM in conjunction with SBM produces additional values. We found cases in which detection of morphological change in either VBM or SBM was superior, and others that showed equivalent performance between the two methods. Therefore, we concluded that using VBM and SBM together can help researchers and clinicians obtain a better understanding of normal neurobiological processes of the brain. Moreover, the use of both methods may improve the accuracy of the detection of morphological changes when comparing the data of patients and controls.In addition, we introduce two other recent methods as future directions for estimating cortical morphological changes: a multi-modal parcellation method using structural and functional images, and a synthetic segmentation method using multi-contrast images (such as T1- and proton density-weighted images).

Using modulated and smoothed data improves detectability of volume difference in group comparison, but reduces accuracy with atlas-based volumetry using Statistical Parametric Mapping 12 software.

BACKGROUND: Atlas-based volumetry using three-dimensional T1-weighted (3D-T1W) magnetic resonance imaging (MRI) has been used previously to evaluate the volumes of intracranial tissues. PURPOSE: To evaluate the detectability of volume difference and accuracy for volumetry using smoothed data with an atlas-based method. MATERIAL AND METHODS: Twenty healthy individuals and 24 patients with idiopathic normal-pressure hydrocephalus (iNPH) underwent 3-T MRI, and sagittal 3D-T1W images were obtained in all participants. Signal values (as tissue probability) of voxels in five segmented data types (gray matter, white matter, cerebrospinal fluid [CSF], skull, soft tissue) derived from the 3D-T1W images with SPM 12 software were assigned simulated 3D-T1W signal intensities to each tissue image. The assigned data were termed "reference data." We created a reference 3D-T1W image that included the reference data of all five tissue types. Standard volumes were measured for the reference CSF data with region of interest of lateral ventricle in native space, and measured volumes were obtained for non-smoothed and smoothed-modulated data. Detectability was evaluated between measured volumes in the healthy control and iNPH groups. Accuracy was evaluated as the difference between the mean measured and standard volumes. RESULTS: In group comparison of measured volumes between the healthy control and iNPH groups, the lowest P value was for smoothed-modulated CSF data. In both groups, the largest difference from the standard volume was found for the mean of the measured volumes for smoothed-modulated CSF data. CONCLUSION: Our study shows that using smoothed data can improve detectability in group comparison. However, using smoothed data reduces the accuracy of volumetry.

A Contribution to the History of Japanese Education Systems for Radiological Technologists.

Background: The evolution of radiological technology is one of the most remarkable events of modern medical technology. Radiological examination has resulted in non-invasive, individual diagnostic imaging, which has contributed significantly to successful medical treatment of patients. Key Concepts: This review summarizes past and current Japanese educational systems for radiological technologists with a historical perspective focusing on three periods. The first period begins with Roentgen's discovery of X-rays (1895), the second period begins with the establishment of the Radiological X-ray Technologist Act (1951), and the third period begins with the launch of the first university course for radiological technologists (1987). It is conceivable that those periods are in accordance with the technological paradigm shifts, including the development of contrast radiography and the application of CT and MRI to clinical practice. To maintain awareness of the most recent available technologies and maximize safety, educational programs teaching the latest knowledge were offered during each period. Conclusions: The advanced technologies require highly skilled radiological technologists and highly established educational systems. At present, over 70% of Japanese educational programs for radiological technologists are university courses leading to a bachelor's degree. The increasing globalization of radiological technology requires future radiological education systems to have a global perspective.

Quantitative evaluation of deep convolutional neural network-based image denoising for low-dose computed tomography.

To minimize radiation risk, dose reduction is important in the diagnostic and therapeutic applications of computed tomography (CT). However, image noise degrades image quality owing to the reduced X-ray dose and a possible unacceptably reduced diagnostic performance. Deep learning approaches with convolutional neural networks (CNNs) have been proposed for natural image denoising; however, these approaches might introduce image blurring or loss of original gradients. The aim of this study was to compare the dose-dependent properties of a CNN-based denoising method for low-dose CT with those of other noise-reduction methods on unique CT noise-simulation images. To simulate a low-dose CT image, a Poisson noise distribution was introduced to normal-dose images while convoluting the CT unit-specific modulation transfer function. An abdominal CT of 100 images obtained from a public database was adopted, and simulated dose-reduction images were created from the original dose at equal 10-step dose-reduction intervals with a final dose of 1/100. These images were denoised using the denoising network structure of CNN (DnCNN) as the general CNN model and for transfer learning. To evaluate the image quality, image similarities determined by the structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) were calculated for the denoised images. Significantly better denoising, in terms of SSIM and PSNR, was achieved by the DnCNN than by other image denoising methods, especially at the ultra-low-dose levels used to generate the 10% and 5% dose-equivalent images. Moreover, the developed CNN model can eliminate noise and maintain image sharpness at these dose levels and improve SSIM by approximately 10% from that of the original method. In contrast, under small dose-reduction conditions, this model also led to excessive smoothing of the images. In quantitative evaluations, the CNN denoising method improved the low-dose CT and prevented over-smoothing by tailoring the CNN model.

Influence of Mild White Matter Lesions on Voxel-based Morphometry.

PURPOSE: The aim of this study was to investigate whether the detectability of brain volume change in voxel-based morphometry (VBM) with gray matter images is affected by mild white matter lesions (MWLs). METHODS: Three-dimensional T1-weighted images (3D-T1WIs) of 11 healthy subjects were obtained using a 3T MR scanner. We initially created 3D-T1WIs with focal cortical atrophy simulated cortical atrophy in left amygdala (type A) and the left medial frontal lobe (type B) from control 3D-T1WIs. Next, the following three types of MWL images were created: type A + 1L and type B + 1L images, only one white matter lesion; type A + 4L and type B + 4L images, four white matter lesions at distant positions; and type A + 4L* and type B + 4L* images, four white matter lesions at clustered positions. Comparisons between the control group and the other groups were performed with VBM using segmented gray matter images. RESULTS: The gray matter volume was significantly lower in the type A group than in the control group, and similar results were observed in the type A + 1L, type A + 4L, and type A + 4L* groups. Additionally, the gray matter volume was significantly lower in the type B group than in the control group, and similar results were observed in the type B + 1L, type B + 4L, and type B + 4L* groups, but the cluster size in type B + 4L* was smaller than that in type B. CONCLUSION: Our study showed that the detectability of brain volume change in VBM with gray matter images was not decreased by MWLs as lacunar infarctions. Therefore, we think that group comparisons with VBM should be analyzed by groups including and excluding subjects with MWLs, respectively.

Estimation of intracranial volume: A comparative study between synthetic MRI and FSL-brain extraction tool (BET)2.

Brain extraction represents an important step in numerous neuroimaging analyses. The brain extraction tool (BET)2 is a widely used deformable model-based approach for extraction of intracranial volume (ICV). The aim of this study is to estimate the ICV extraction accuracy using synthetic MR(SyMRI) method and BET2 in healthy adult participants and patients with Sturge-Weber Syndrome (SWS), including infants. 'Quantification of relaxation times and proton density by multi-echo acquisition of saturation recovery with turbo-spin-echo readout' (QRAPMASTER) with a 3.0 T magnetic resonance image (MRI) system was used for data acquisition. Statistical evaluations were performed with linear regression analysis and the Jaccard similarity coefficient (J). ICV extraction accuracy with synthetic MR method is found to be higher than BET2, for both aged healthy participants and SWS.

Effect of changing the analyzed image contrast on the accuracy of intracranial volume extraction using Brain Extraction Tool 2.

The aim of this study was to evaluate the effect of changing the contrast of an analyzed image on the accuracy of intracranial volume (ICV) extraction using the Brain Extraction Tool (BET2) in healthy adults and patients with Sturge-Weber syndrome (SWS), including infants. Twelve SWS patients, including infants, and 12 healthy participants were imaged on a 3.0-T magnetic resonance imaging (MRI) machine. All individuals underwent quantification of relaxation times and proton density using multi-echo acquisition of saturation recovery with turbo-spin-echo readout (QRAPMASTER). Based on the QRAPMASTER data, we created images with seven contrasts (T1-WI, T2-WI, PD-WI, T2 short-tau inversion recovery [STIR], proton density [PD] STIR, T2STIR + PDSTIR, and T1-WI + T2-WI + PD-WI) by post-processing with SyMRI software. ICVs extracted with BET2 from the FMRIB (Functional Magnetic Resonance Imaging of the Brain) Software Library with each of the seven image contrasts were compared with manually extracted ICVs, which is the gold standard reviewed by a board-certificated neuroradiologist. Manual extraction was performed on T1-WI and T2STIR. Statistical analyses were performed with Jaccard similarity coefficients (J). The highest J score was found in T1-WI + T2-WI + PD-WI in all participants (0.8451); T1-WI in healthy participants (0.8984); T2STIR in participants with SWS (0.8325). Our findings suggest that T1-WI and T2STIR should be used in ICV extraction performed using BET2 on healthy participants and infants, respectively. Additionally, if the analyzed individuals include both healthy participants and infants, T1-WI + T2-WI + PD-WI should be used.