Search

Background To accurately diagnose and control heart failure (HF), it is important to carry out a simple assessment of elevated pulmonary arterial wedge pressure (PAWP). The aim of this study was to develop and validate an objective method for detecting elevated PAWP by applying deep learning (DL) to a chest x-ray (CXR). Methods We enrolled 1013 consecutive patients with a right-heart catheter between October 2009 and February 2020. We developed a convolutional neural network to identify patients with elevated PAWP (> 18 mm Hg) as the actual value of PAWP to be used in the dataset for training. In the prospective validation dataset used to detect elevated PAWP, the area under the receiver operating characteristic curve (AUC) was calculated using the DL model that evaluated the CXR. Results In the prospective validation dataset, the AUC of the DL model with CXR was not significantly different from the AUC produced by brain natriuretic peptide (BNP) and the echocardiographic left-ventricular diastolic dysfunction (DD) algorithm (DL model: 0.77 vs BNP: 0.77 vs DD algorithm: 0.70; respectively; P = NS for all comparisons); it was, however, significantly higher than the AUC of the cardiothoracic ratio (DL model vs cardiothoracic ratio [CTR]: 0.66, P = 0.044). The model based on 3 parameters (BNP, DD algorithm, and CTR) was improved by adding the DL model (AUC: from 0.80 to 0.86; P = 0.041). Conclusions Applying the DL model based on a CXR (a classical, universal, and low-cost test) is useful for screening for elevated PAWP.

Stem cell‐based products have clinical and industrial applications. Thus, there is a need to develop quality control methods to standardize stem cell manufacturing. Here, we report a deep learning‐based automated cell tracking (DeepACT) technology for noninvasive quality control and identification of cultured human stem cells. The combination of deep learning‐based cascading cell detection and Kalman filter algorithm‐based tracking successfully tracked the individual cells within the densely packed human epidermal keratinocyte colonies in the phase‐contrast images of the culture. DeepACT rapidly analyzed the motion of individual keratinocytes, which enabled the quantitative evaluation of keratinocyte dynamics in response to changes in culture conditions. Furthermore, DeepACT can distinguish keratinocyte stem cell colonies from non‐stem cell‐derived colonies by analyzing the spatial and velocity information of cells. This system can be widely applied to stem cell cultures used in regenerative medicine and provides a platform for developing reliable and noninvasive quality control technology., A deep learning‐based automated cell tracking (DeepACT) technology enables the evaluation of keratinocyte culture quality and the identification of keratinocyte stem cells using quantitative cell motion analysis. DeepACT comprises two main modules: identifying human keratinocytes at single‐cell resolution from phase‐contrast images of cultures through deep learning and tracking keratinocyte motion in the colony using a state‐space model.

Accurate diagnosis of pulmonary hypertension (PH) is crucial to ensure that patients receive timely treatment. We hypothesized that application of artificial intelligence (AI) to the chest X-ray (CXR) could identify elevated pulmonary artery pressure (PAP) and stratify the risk of heart failure hospitalization with PH. We retrospectively enrolled a total of 900 consecutive patients with suspected PH. We trained a convolutional neural network to identify patients with elevated PAP (> 20 mmHg) as the actual value of PAP. The endpoints in this study were admission or occurrence of heart failure with elevated PAP. In an independent evaluation set for detection of elevated PAP, the area under curve (AUC) by the AI algorithm was significantly higher than the AUC by measurements of CXR images and human observers (0.71 vs. 0.60 and vs. 0.63, all p

The purpose of this study was to evaluate the accuracy of the convolutional neural network (CNN) model in glaucoma identification with three primary colors (red, green, blue; RGB) and split color channels using fundus photographs with a small sample size. The dataset was prepared using color fundus photographs captured with a fundus camera (VX-10i, Kowa Co., Ltd., Tokyo, Japan). The training dataset consisted of 200 images, and the validation dataset contained 60 images. In the preprocessing stage, the color channels for the fundus images were separated into red (red model), green (green model), and blue (blue model) using OpenCV on Windows. All images were resized to squares with a size of 512 × 512 pixels for preprocessing before input into the model, and the model was fine-tuned with VGG16. The diagnostic performance was significantly higher in the green model [area under the curve (AUC) 0.946; 95% confidence interval (CI) 0.851–0.982] than in the RGB model (AUC 0.800; 95% CI 0.658–0.893; P = 0.006), red model (AUC 0.746; 95% CI 0.601–0.851; P = 0.002), and blue model (AUC 0.558; 95% CI 0.405–0.700; P

Background Classification of optical coherence tomography (OCT) images can be achieved with high accuracy using classical convolution neural networks (CNN), a commonly used deep learning network for computer-aided diagnosis. Classical CNN has often been criticized for suppressing positional relations in a pooling layer. Therefore, because capsule networks can learn positional information from images, we attempted application of a capsule network to OCT images to overcome that shortcoming. This study is our attempt to improve classification accuracy by replacing CNN with a capsule network. Methods From an OCT dataset, we produced a training dataset of 83,484 images and a test dataset of 1000 images. For training, the dataset comprises 37,205 images with choroidal neovascularization (CNV), 11,348 with diabetic macular edema (DME), 8616 with drusen, and 26,315 normal images. The test dataset has 250 images from each category. The proposed model was constructed based on a capsule network for improving classification accuracy. It was trained using the training dataset. Subsequently, the test dataset was used to evaluate the trained model. Results Classification of OCT images using our method achieved accuracy of 99.6%, which is 3.2 percentage points higher than that of other methods described in the literature. Conclusion The proposed method achieved classification accuracy results equivalent to those reported for other methods for CNV, DME, drusen, and normal images.

For interventional radiology, dose management has persisted as a crucially important issue to reduce radiation exposure to patients and medical staff. This study designed a real-time dose visualization system for interventional radiology designed with mixed reality technology and Monte Carlo simulation. An earlier report described a Monte-Carlo-based estimation system, which simulates a patient’s skin dose and air dose distributions, adopted for our system. We also developed a system of acquiring fluoroscopic conditions to input them into the Monte Carlo system. Then we combined the Monte Carlo system with a wearable device for three-dimensional holographic visualization. The estimated doses were transferred sequentially to the device. The patient’s dose distribution was then projected on the patient body. The visualization system also has a mechanism to detect one’s position in a room to estimate the user’s exposure dose to detect and display the exposure level. Qualitative tests were conducted to evaluate the workload and usability of our mixed reality system. An end-to-end system test was performed using a human phantom. The acquisition system accurately recognized conditions that were necessary for real-time dose estimation. The dose hologram represents the patient dose. The user dose was changed correctly, depending on conditions and positions. The perceived overall workload score (33.50) was lower than the scores reported in the literature for medical tasks (50.60) for computer activities (54.00). Mixed reality dose visualization is expected to improve exposure dose management for patients and health professionals by exhibiting the invisible radiation exposure in real space. Supplementary Information The online version of this article (10.1007/s10916-020-01700-9) contains supplementary material, which is available to authorized users.

Out‐of‐field organs are not commonly designated as dose calculation targets during radiation therapy treatment planning, but they might entail risks of second cancer. Risk components include specific internal body scatter, which is a dominant source of out‐of‐field doses, and head leakage, which can be reduced by external shielding. Our simulation study quantifies out‐of‐field organ doses and estimates second cancer risks attributable to internal body scatter in whole‐breast radiotherapy (WBRT) with or without additional regional nodal radiotherapy (RNRT), respectively, for right and left breast cancer using Monte Carlo code PHITS. Simulations were conducted using a complete whole‐body female model. Second cancer risk was estimated using the calculated doses with a concept of excess absolute risk. Simulation results revealed marked differences between WBRT alone and WBRT plus RNRT in out‐of‐field organ doses. The ratios of mean doses between them were as large as 3.5–8.0 for the head and neck region and about 1.5–6.6 for the lower abdominal region. Potentially, most out‐of‐field organs had excess absolute risks of less than 1 per 10,000 persons‐year. Our study surveyed the respective contributions of internal body scatter to out‐of‐field organ doses and second cancer risks in breast radiotherapy on this intact female model.

This study was conducted to improve cone-beam computed tomography (CBCT) image quality using the super-resolution technique, a method of inferring a high-resolution image from a low-resolution image. This technique is used with two matrices, so-called dictionaries, constructed respectively from high-resolution and low-resolution image bases. For this study, a CBCT image, as a low-resolution image, is represented as a linear combination of atoms, the image bases in the low-resolution dictionary. The corresponding super-resolution image was inferred by multiplying the coefficients and the high-resolution dictionary atoms extracted from planning CT images. To evaluate the proposed method, we computed the root mean square error (RMSE) and structural similarity (SSIM). The resulting RMSE and SSIM between the super-resolution images and the planning CT images were, respectively, as much as 0.81 and 1.29 times better than those obtained without using the super-resolution technique. We used super-resolution technique to improve the CBCT image quality.

Low-count SPECT images are well known to be smoothed strongly by a Butterworth filter for statistical noise reduction. Reconstructed images have a low signal-to-noise ratio (SNR) and spatial resolution because of the removal of high-frequency signal components. Using the developed robust adaptive bilateral filter (RABF), which was designed as a pre-stage filter of the Butterworth filter, this study was conducted to improve SNR without degrading the spatial resolution for low-count SPECT imaging. The filter can remove noise while preserving spatial resolution. To evaluate the proposed method, we extracted SNR and spatial resolution in a phantom study. We also conducted paired comparison for visual image quality evaluation in a clinical study. Results show that SNR was increased 1.4 times without degrading the spatial resolution. Visual image quality was improved significantly (p < 0.01) for clinical low-count data. Moreover, the accumulation structure became sharper. A structure embedded in noise emerged. Our method, which denoises without degrading the spatial resolution for low-count SPECT images, is expected to increase the effectiveness of diagnosis for low-dose scanning and short acquisition time scanning.

To minimise the radiation dermatitis related to interventional radiology (IR), rapid and accurate dose estimation has been sought for all procedures. We propose a technique for estimating the patient skin dose rapidly and accurately using Monte Carlo (MC) simulation with a graphical processing unit (GPU, GTX 1080; Nvidia Corp.). The skin dose distribution is simulated based on an individual patient’s computed tomography (CT) dataset for fluoroscopic conditions after the CT dataset has been segmented into air, water and bone based on pixel values. The skin is assumed to be one layer at the outer surface of the body. Fluoroscopic conditions are obtained from a log file of a fluoroscopic examination. Estimating the absorbed skin dose distribution requires calibration of the dose simulated by our system. For this purpose, a linear function was used to approximate the relation between the simulated dose and the measured dose using radiophotoluminescence (RPL) glass dosimeters in a water-equivalent phantom. Differences of maximum skin dose between our system and the Particle and Heavy Ion Transport code System (PHITS) were as high as 6.1%. The relative statistical error (2 σ) for the simulated dose obtained using our system was ≤3.5%. Using a GPU, the simulation on the chest CT dataset aiming at the heart was within 3.49 s on average: the GPU is 122 times faster than a CPU (Core i7–7700K; Intel Corp.). Our system (using the GPU, the log file, and the CT dataset) estimated the skin dose more rapidly and more accurately than conventional methods.

The purpose of this study is to evaluate the accuracy for classification of hepatic tumors by characterization of T1-weighted magnetic resonance (MR) images using two radiomics approaches with machine learning models: texture analysis and topological data analysis using persistent homology. This study assessed non-contrast-enhanced fat-suppressed three-dimensional (3D) T1-weighted images of 150 hepatic tumors. The lesions included 50 hepatocellular carcinomas (HCCs), 50 metastatic tumors (MTs), and 50 hepatic hemangiomas (HHs) found respectively in 37, 23, and 33 patients. For classification, texture features were calculated, and also persistence images of three types (degree 0, degree 1 and degree 2) were obtained for each lesion from the 3D MR imaging data. We used three classification models. In the classification of HCC and MT (resp. HCC and HH, HH and MT), we obtained accuracy of 92% (resp. 90%, 73%) by texture analysis, and the highest accuracy of 85% (resp. 84%, 74%) when degree 1 (resp. degree 1, degree 2) persistence images were used. Our methods using texture analysis or topological data analysis allow for classification of the three hepatic tumors with considerable accuracy, and thus might be useful when applied for computer-aided diagnosis with MR images.

To identify predictive factors for embolic material conversion to N-butyl cyanoacrylate (NBCA) for the treatment of primary postpartum hemorrhage (PPH) after failed transcatheter arterial embolization (TAE) using gelatin sponge (GS). Institutional review board approval was obtained. We retrospectively studied 62 consecutive women with primary PPH who underwent TAE between January 2006 and March 2015. Five of them were excluded for the following: cardiopulmonary arrest at arrival (n = 1), uterine inversion (n = 1), and hysterectomy after TAE (n = 3). Remaining 57 women (age range, 21–43 years; mean, 32.6 years) comprised study population. TAE was initially performed using GS in all cases and then converted to NBCA after two embolizations using GS with persistent hemodynamic instability or vaginal bleeding. The patients’ background, uterine height, vital signs, laboratory tests, disseminated intravascular coagulation score, and details of procedure were reviewed. Univariate and multivariate analyses were performed to determine factors related to embolic material conversion. Technical success rate was 100%. Fourteen patients (25%) needed embolic material conversion to NBCA. Univariate analysis showed that uterine height, systolic blood pressure (sBP), and hemoglobin level were significantly related to embolic material conversion to NBCA (P = 0.029, 0.030, and 0.042). Logistic regression analysis showed that uterine height (odds ratio, 1.37; P = 0.025) and sBP (odds ratio, 0.96; P = 0.003) were associated with embolic material conversion to NBCA. Uterine height and sBP can be predictive factors for embolic material conversion to NBCA for the treatment of PPH. Level 4, Case Control Study

The realization of automatic anomaly detection of respiratory motion could be very useful to prevent accidental damage during radiation therapy. In this paper, we proposed an automatic anomaly detection method using singular value decomposition analysis. Before applying this method, the investigator needs a normal respiratory motion data of a patient. From these data, a trajectory matrix representing normal time-series feature is created. Decomposing the matrix, we obtained the feature of normal time series. Then, we applied the same procedure to real-time data and obtained real-time features. Calculating the similarity of those feature matrixes, an anomaly score was obtained. Patient motion was observed by a depth camera. In our simulation, two types of motion e.g. cough and sudden stop of breathing were successfully detected, while gradual change of respiratory cycle frequency was not detected clearly.

In medical imaging, precise and reliable images are very important. However, the quality of medical images is sometimes limited by low-event statistics owing to the low sensitivity of the detectors commonly used in radiology. On the other hand, long exposure to radiation and long inspection duration can become a burden for patients. In this paper, we propose a method for generating high-quality images of gamma ray sources from low statistic data by using machine learning methods based on dictionary learning and sparse coding. As the first application, we generated a high-quality image of 137Cs, which emits 662-keV gamma rays, from low-event statistics measured using a Compton camera. We simulated with Geant4 various geometries of the gamma-ray source (137Cs; 662 keV) as measured with a Compton camera by Geant4. Then, complete sets of low-resolution and high-resolution dictionaries were prepared. We generated super-resolution images from low-resolution test images obtained from actual measurements. The convergence of the gamma-ray images was similar for both the ground truth and predicted images, as supported by the improvements in the structural similarity (SSIM), peak signal-to-noise (PSNR) ratio, and root mean square error (RMSE) in the corresponding images. We also discuss future plans to use the super-resolution technique for visualizing radium chloride (223RaCl2) in the patient’s body, which will make it possible to achieve in-vivo imaging of alpha-particle internal therapy for the first time.

Introduction Cone beam computed tomography (CBCT) plays an important role in image-guided radiation therapy (IGRT), while having disadvantages of severe shading artifact caused by the reconstruction using scatter contaminated and truncated projections. The purpose of this study is to develop a deep convolutional neural network (DCNN) method for improving CBCT image quality. Methods CBCT and planning computed tomography (pCT) image pairs from 20 prostate cancer patients were selected. Subsequently, each pCT volume was pre-aligned to the corresponding CBCT volume by image registration, thereby leading to registered pCT data (pCTr). Next, a 39-layer DCNN model was trained to learn a direct mapping from the CBCT to the corresponding pCTr images. The trained model was applied to a new CBCT data set to obtain improved CBCT (i-CBCT) images. The resulting i-CBCT images were compared to pCTr using the spatial non-uniformity (SNU), the peak-signal-to-noise ratio (PSNR) and the structural similarity index measure (SSIM). Results The image quality of the i-CBCT has shown a substantial improvement on spatial uniformity compared to that of the original CBCT, and a significant improvement on the PSNR and the SSIM compared to that of the original CBCT and the enhanced CBCT by the existing pCT-based correction method. Conclusion We have developed a DCNN method for improving CBCT image quality. The proposed method may be directly applicable to CBCT images acquired by any commercial CBCT scanner.

This phantom study assessed the effect of Gd-EOB-DTPA on T1 bias (difference in T1 between water and fat) of the proton density fat fraction when using magnetic resonance spectroscopy. Phantoms containing varying fat percentages, without and with Gd-EOB-DTPA (precontrast and postcontrast, respectively), were scanned with repetition times ranging from 1000 to 5000 ms. The relationship between the proton density fat fraction at a reference repetition time of 5000 ms and that using different repetition times, was evaluated in the precontrast and postcontrast phantoms using linear regression and Bland–Altman analyses. In the precontrast phantom, as the repetition time increased, the slope tended to approach one. In the postcontrast phantom, the slope and intercept were near one and zero, respectively. The mean difference was smaller in the postcontrast phantom (range − 0.24 to − 0.01%) than in the precontrast phantom (range 0.12 to 3.52%). We conclude that I1 bias is minimized by Gd-EOB-DTPA.

Purpose To evaluate the reproducibility of proton density fat fraction (PDFF) and T2* in a fat-water phantom on three different 3 T MRI systems using 6-point DIXON methods. Methods A phantom which included varying fat volume percentages (true fat fraction [FF]) was scanned by three 3 T MR machines, and PDFF and T2* were measured. Results The mean difference between true FF and PDFF was small in all vendors (−2.11% to 0.41%). However, the difference ratio for T2* values was large among vendors (1.79 to 3.36). Conclusions True FF and PDFF were consistent across vendors; however, T2* varied greatly.

The purpose of this study was to evaluate whether disodium gadoxetate (Gd-EOB-DTPA) affects proton density fat fractions (PDFFs) during use of the multiecho Dixon (meDixon) method in phantom and simulation magnetic resonance imaging (MRI) studies at 3 T. Fat–water phantoms comprising vegetable fat–water emulsions with varying fat volume percentages (0, 5, 10, 15, 20, 30, 40, and 50) and Gd-EOB-DTPA concentrations (0 and 0.4 mM) were prepared. Phantoms without Gd-EOB-DTPA were defined as precontrast, and those with Gd-EOB-DTPA were defined as postcontrast. All phantoms were scanned with a 3 T MRI system using the meDixon method, and precontrast and postcontrast PDFFs were calculated. Simulated pre and postcontrast PDFFs in the liver were calculated using a theoretical formula. The relationship between PDFFs measured in the pre and postcontrast phantoms was evaluated using linear regression and Bland–Altman analysis. The regression analysis comparing the pre and postcontrast PDFFs yielded a slope of 0.77 (P

This study aimed to evaluate (1) the agreement between the true fat fraction (FF) and proton density FF (PDFF) measured using a six-echo modified Dixon (6mDixon) and magnetic resonance spectroscopy (MRS) and (2) the influence of fat on T2* values. The study was performed using phantoms of varying fat and iron content. Point-resolved spectroscopy (PRESS) and stimulated echo acquisition mode (STEAM) with single-echo (S) and multiecho (M) (PRESS-S, PRESS-M, STEAM-S, and STEAM-M) were used for MRS. In phantoms without iron, the agreement between the true FF and measured PDFF was tested using Bland–Altman analysis. The influence of iron on PDFF was evaluated in phantoms with iron. The relationship between the true FF and T2* value was assessed in phantoms without iron, wherein the mean differences (limits of agreement) for each method were as follows: 6mDixon 2.9% (−2.4 to 8.1%); STEAM-S 3.2% (−9.5 to 16.0%); STEAM-M −0.7% (−6.9 to 5.5%); PRESS-S 8.9% (−14.5 to 32.4%); and PRESS-M −5.8% (−18.3 to 6.7%). In the 20% fat phantoms with iron, as iron increased, PDFFs with STEAM-S, PRESS-S, and PRESS-M were considerably overestimated, while, PDFF with STEAM-M was stable at 0.04–0.2 mM iron concentrations (17.2 and 21.4%, respectively), and PDFF with 6mDixon was reliable at even 0.4 mM iron concentration (24.8%). The T2* value showed a negative correlation with the true FF (r = −0.942, P = 0.005). STEAM-M and 6mDixon were reliable methods of fat quantification in the absence of iron, and the T2* value was shortened by fat.

Purpose The purpose of this study is to introduce the new concept of a four-dimensional (4D) cone-beam computed tomography (CBCT) reconstruction approach for non-periodic organ motion in cooperation with the time-ordered chain graph model (TCGM) and to compare it with previously developed methods such as total variation-based compressed sensing (TVCS) and prior-image constrained compressed sensing (PICCS). Materials and Methods Our proposed reconstruction is based on a model including the constraint originating from the images of neighboring time phases. Namely, the reconstructed time-series images depend on each other in this TCGM scheme, and the time-ordered images are concurrently reconstructed in the iterative reconstruction approach. In this study, iterative reconstruction with the TCGM was carried out with 90° projection ranges. The images reconstructed by the TCGM were compared with the images reconstructed by TVCS (200° projection ranges) and PICCS (90° projection ranges). Two kinds of projection data sets–an elliptic-cylindrical digital phantom and two clinical patients’ data–were used. For the digital phantom, an air sphere was contained and virtually moved along the longitudinal axis by 3 cm/30 s and 3 cm/60 s; the temporal resolution was evaluated by measuring the penumbral width of the air sphere. The clinical feasibility of the non-periodic time-ordered 4D CBCT image reconstruction was examined with the patient data in the pelvic region. Results In the evaluation of the digital-phantom reconstruction, the penumbral widths of the TCGM yielded the narrowest result; the results obtained by PICCS and TCGM using 90° projection ranges were 2.8% and 18.2% for 3 cm/30 s, and 5.0% and 23.1% for 3 cm/60 s narrower than that of TVCS using 200° projection ranges. This suggests that the TCGM has a better temporal resolution, whereas PICCS seems similar to TVCS. These reconstruction methods were also compared using patients’ projection data sets. Although all three reconstruction results showed motion related to rectal gas or stool, the result obtained by the TCGM was visibly clearer with less blurring. Conclusion The TCGM is a feasible approach to visualize non-periodic organ motion. The digital-phantom results demonstrated that the proposed method provides 4D image series with a better temporal resolution compared to TVCS and PICCS. The clinical patients’ results also showed that the present method enables us to visualize motion related to rectal gas and flatus in the rectum.