Search

Purpose This study aimed to enhance the multidimensional nominal response model (MDNRM) for multiclass classification in diagnostic radiology. Materials and Methods This retrospective study involved the extension of the conventional nominal response model (NRM) to create the two-parameter MDNRM (2PL-MDNRM). Seven models of MDNRM, including the original MDNRM and subtypes of 2PL-MDNRM, were employed to estimate test-takers’ abilities and test item complexity. These models were applied to a clinical diagnostic radiology dataset. Rhat values were calculated to evaluate model convergence. Additionally, values of the widely applicable information criterion (wAIC) and Pareto-smoothed importance sampling leave-one-out cross-validation (LOO) were calculated to evaluate the goodness of fit of the seven models. The best-performing model was selected based on the values of wAIC and LOO. Probability of direction (PD) was used to evaluate whether one estimated parameter significantly differed. Results All estimated parameters across the seven models demonstrated Rhat values below 1.10, indicating stable convergence. The best wAIC and LOO values (988 and 1,121, respectively) were achieved with 2PL-MDNRMr using the truncated normal distribution and 2PL-MDNRMa using the truncated normal distribution. Notably, one test-taker (radiologist) exhibited significantly superior ability compared to another based on PD results from the best models, while no significant difference was observed in nonoptimal models. Conclusion 2PL-MDNRM successfully achieved parameter estimation convergence, and its superiority over the original MDNRM was demonstrated through wAIC and LOO values.

Rationale and objectives: To develop and validate a deep learning (DL) model to automatically diagnose muscle-invasive bladder cancer (MIBC) on MRI with Vision Transformer (ViT). Materials and methods: This multicenter retrospective study included patients with BC who reported to two institutions between January 2016 and June 2020 (training dataset) and a third institution between May 2017 and May 2022 (test dataset). The diagnostic model for MIBC and the segmentation model for BC on MRI were developed using the training dataset with 5-fold cross-validation. ViT- and convolutional neural network (CNN)-based diagnostic models were developed and compared for diagnostic performance using the area under the curve (AUC). The performance of the diagnostic model with manual and auto-generated regions of interest (ROImanual and ROIauto, respectively) was validated on the test dataset and compared to that of radiologists (three senior and three junior radiologists) using Vesical Imaging Reporting and Data System scoring. Results: The training and test datasets included 170 and 53 patients, respectively. Mean AUC of the top 10 ViT-based models with 5-fold cross-validation outperformed those of the CNN-based models (0.831 ± 0.003 vs. 0.713 ± 0.007–0.812 ± 0.006, p

Abstract To evaluate the diagnostic performance of our deep learning (DL) model of COVID-19 and investigate whether the diagnostic performance of radiologists was improved by referring to our model. Our datasets contained chest X-rays (CXRs) for the following three categories: normal (NORMAL), non-COVID-19 pneumonia (PNEUMONIA), and COVID-19 pneumonia (COVID). We used two public datasets and private dataset collected from eight hospitals for the development and external validation of our DL model (26,393 CXRs). Eight radiologists performed two reading sessions: one session was performed with reference to CXRs only, and the other was performed with reference to both CXRs and the results of the DL model. The evaluation metrics for the reading session were accuracy, sensitivity, specificity, and area under the curve (AUC). The accuracy of our DL model was 0.733, and that of the eight radiologists without DL was 0.696 ± 0.031. There was a significant difference in AUC between the radiologists with and without DL for COVID versus NORMAL or PNEUMONIA (p = 0.0038). Our DL model alone showed better diagnostic performance than that of most radiologists. In addition, our model significantly improved the diagnostic performance of radiologists for COVID versus NORMAL or PNEUMONIA.

Background and purposeMean pulmonary artery pressure (mPAP) is a key index for chronic thromboembolic pulmonary hypertension (CTEPH). Using machine learning, we attempted to construct an accurate prediction model for mPAP in patients with CTEPH.MethodsA total of 136 patients diagnosed with CTEPH were included, for whom mPAP was measured. The following patient data were used as explanatory variables in the model: basic patient information (age and sex), blood tests (brain natriuretic peptide (BNP)), echocardiography (tricuspid valve pressure gradient (TRPG)), and chest radiography (cardiothoracic ratio (CTR), right second arc ratio, and presence of avascular area). Seven machine learning methods including linear regression were used for the multivariable prediction models. Additionally, prediction models were constructed using the AutoML software. Among the 136 patients, 2/3 and 1/3 were used as training and validation sets, respectively. The average of R squared was obtained from 10 different data splittings of the training and validation sets.ResultsThe optimal machine learning model was linear regression (averaged R squared, 0.360). The optimal combination of explanatory variables with linear regression was age, BNP level, TRPG level, and CTR (averaged R squared, 0.388). The R squared of the optimal multivariable linear regression model was higher than that of the univariable linear regression model with only TRPG.ConclusionWe constructed a more accurate prediction model for mPAP in patients with CTEPH than a model of TRPG only. The prediction performance of our model was improved by selecting the optimal machine learning method and combination of explanatory variables.

Purpose: Natural language processing using language models has yielded promising results in various fields. Language models can help improve the workflow of radiologists. This retrospective study aimed to construct and evaluate language models for automatic summarization of radiology reports. Methods: Two radiology report datasets from the MIMIC Chest X-ray (MIMIC-CXR) database and the Japan Medical Image Database (JMID) were included in this study. The MIMIC-CXR is an open database comprising chest radiograph reports. The JMID is a large database comprising computed tomography and magnetic resonance imaging reports from 10 academic medical centers in Japan. A total of 128,032 and 1,101,271 reports were included in this study from the MIMIC-CXR database and JMID, respectively. Four Text-to-Text Transfer Transformer (T5) models were constructed. Recall-Oriented Understudy for Gisting Evaluation (ROUGE), a quantitative metric, was used to evaluate the quality of the text summarized from 19,205 and 58,043 test sets from the MIMIC-CXR and JMID, respectively. The Wilcoxon signed-rank test was used to evaluate the differences among the ROUGE values of the four T5 models. Moreover, the subsets of automatically summarized text in the test sets were manually evaluated by two radiologists. The best T5 models were selected for automatic summarization using the Wilcoxon signed-rank test. Results: The quantitative metrics of the best T5 models were as follows: ROUGE-1 = 57.75 ± 30.99, ROUGE-2 = 49.96 ± 35.36, and ROUGE-L = 54.07 ± 32.48 in the MIMIC-CXR; and ROUGE-1 = 50.00 ± 29.24, ROUGE-2 = 39.66 ± 30.21, and ROUGE-L = 47.87 ± 29.44 in the JMID. The radiologists’ evaluations revealed 86% and 85% of the texts automatically summarized from the MIMIC-CXR and JMID, respectively, to be clinically useful. Conclusion: The T5 models constructed in this study were able to perform automatic summarization of the radiology reports. The radiologists’ evaluations demonstrated most of the automatically summarized texts to be clinically valuable.

PurposeTo examine the molecular biological differences between conjunctival mucosa-associated lymphoid tissue (MALT) lymphoma and orbital MALT lymphoma in ocular adnexa lymphoma.MethodsObservational case series. A total of 129 consecutive, randomized cases of ocular adnexa MALT lymphoma diagnosed histopathologically between 2008 and 2020.Total RNA was extracted from formalin-fixed paraffin-embedded tissue from ocular adnexa MALT lymphoma, and RNA-sequencing was performed. Orbital MALT lymphoma gene expression was compared with that of conjunctival MALT lymphoma. Gene set (GS) analysis detecting for gene set cluster was performed in RNA-sequence. Related proteins were further examined by immunohistochemical staining. In addition, artificial segmentation image used to count stromal area in HE images.ResultsGS analysis showed differences in expression in 29 GS types in primary orbital MALT lymphoma (N=5,5, FDR q-value 1) independent of age and sex on multivariate analysis (P=0.004). Stroma area in tumor were significant difference(P

Purpose The purpose of this study is to compare two libraries dedicated to the Markov chain Monte Carlo method: pystan and numpyro. In the comparison, we mainly focused on the agreement of estimated latent parameters and the performance of sampling using the Markov chain Monte Carlo method in Bayesian item response theory (IRT). Materials and methods Bayesian 1PL-IRT and 2PL-IRT were implemented with pystan and numpyro. Then, the Bayesian 1PL-IRT and 2PL-IRT were applied to two types of medical data obtained from a published article. The same prior distributions of latent parameters were used in both pystan and numpyro. Estimation results of latent parameters of 1PL-IRT and 2PL-IRT were compared between pystan and numpyro. Additionally, the computational cost of the Markov chain Monte Carlo method was compared between the two libraries. To evaluate the computational cost of IRT models, simulation data were generated from the medical data and numpyro. Results For all the combinations of IRT types (1PL-IRT or 2PL-IRT) and medical data types, the mean and standard deviation of the estimated latent parameters were in good agreement between pystan and numpyro. In most cases, the sampling time using the Markov chain Monte Carlo method was shorter in numpyro than that in pystan. When the large-sized simulation data were used, numpyro with a graphics processing unit was useful for reducing the sampling time. Conclusion Numpyro and pystan were useful for applying the Bayesian 1PL-IRT and 2PL-IRT. Our results show that the two libraries yielded similar estimation result and that regarding to sampling time, the fastest libraries differed based on the dataset size.

Abstract This study aimed to develop a versatile automatic segmentation model of bladder cancer (BC) on MRI using a convolutional neural network and investigate the robustness of radiomics features automatically extracted from apparent diffusion coefficient (ADC) maps. This two-center retrospective study used multi-vendor MR units and included 170 patients with BC, of whom 140 were assigned to training datasets for the modified U-net model with five-fold cross-validation and 30 to test datasets for assessment of segmentation performance and reproducibility of automatically extracted radiomics features. For model input data, diffusion-weighted images with b = 0 and 1000 s/mm2, ADC maps, and multi-sequence images (b0-b1000-ADC maps) were used. Segmentation accuracy was compared between ours and existing models. The reproducibility of radiomics features on ADC maps was evaluated using intraclass correlation coefficient. The model with multi-sequence images achieved the highest Dice similarity coefficient (DSC) with five-fold cross-validation (mean DSC = 0.83 and 0.79 for the training and validation datasets, respectively). The median (interquartile range) DSC of the test dataset model was 0.81 (0.70–0.88). Radiomics features extracted from manually and automatically segmented BC exhibited good reproducibility. Thus, our U-net model performed highly accurate segmentation of BC, and radiomics features extracted from the automatic segmentation results exhibited high reproducibility.

Abstract The integrated positron emission tomography/magnetic resonance imaging (PET/MRI) scanner simultaneously acquires metabolic information via PET and morphological information using MRI. However, attenuation correction, which is necessary for quantitative PET evaluation, is difficult as it requires the generation of attenuation-correction maps from MRI, which has no direct relationship with the gamma-ray attenuation information. MRI-based bone tissue segmentation is potentially available for attenuation correction in relatively rigid and fixed organs such as the head and pelvis regions. However, this is challenging for the chest region because of respiratory and cardiac motions in the chest, its anatomically complicated structure, and the thin bone cortex. We propose a new method using unsupervised generative attentional networks with adaptive layer-instance normalisation for image-to-image translation (U-GAT-IT), which specialised in unpaired image transformation based on attention maps for image transformation. We added the modality-independent neighbourhood descriptor (MIND) to the loss of U-GAT-IT to guarantee anatomical consistency in the image transformation between different domains. Our proposed method obtained a synthesised computed tomography of the chest. Experimental results showed that our method outperforms current approaches. The study findings suggest the possibility of synthesising clinically acceptable computed tomography images from chest MRI with minimal changes in anatomical structures without human annotation.

Abstract This retrospective study aimed to develop and validate a deep learning model for the classification of coronavirus disease-2019 (COVID-19) pneumonia, non-COVID-19 pneumonia, and the healthy using chest X-ray (CXR) images. One private and two public datasets of CXR images were included. The private dataset included CXR from six hospitals. A total of 14,258 and 11,253 CXR images were included in the 2 public datasets and 455 in the private dataset. A deep learning model based on EfficientNet with noisy student was constructed using the three datasets. The test set of 150 CXR images in the private dataset were evaluated by the deep learning model and six radiologists. Three-category classification accuracy and class-wise area under the curve (AUC) for each of the COVID-19 pneumonia, non-COVID-19 pneumonia, and healthy were calculated. Consensus of the six radiologists was used for calculating class-wise AUC. The three-category classification accuracy of our model was 0.8667, and those of the six radiologists ranged from 0.5667 to 0.7733. For our model and the consensus of the six radiologists, the class-wise AUC of the healthy, non-COVID-19 pneumonia, and COVID-19 pneumonia were 0.9912, 0.9492, and 0.9752 and 0.9656, 0.8654, and 0.8740, respectively. Difference of the class-wise AUC between our model and the consensus of the six radiologists was statistically significant for COVID-19 pneumonia (p value = 0.001334). Thus, an accurate model of deep learning for the three-category classification could be constructed; the diagnostic performance of our model was significantly better than that of the consensus interpretation by the six radiologists for COVID-19 pneumonia.

Abstract To determine whether temporal subtraction (TS) CT obtained with non-rigid image registration improves detection of various bone metastases during serial clinical follow-up examinations by numerous radiologists. Six board-certified radiologists retrospectively scrutinized CT images for patients with history of malignancy sequentially. These radiologists selected 50 positive and 50 negative subjects with and without bone metastases, respectively. Furthermore, for each subject, they selected a pair of previous and current CT images satisfying predefined criteria by consensus. Previous images were non-rigidly transformed to match current images and subtracted from current images to automatically generate TS images. Subsequently, 18 radiologists independently interpreted the 100 CT image pairs to identify bone metastases, both without and with TS images, with each interpretation separated from the other by an interval of at least 30 days. Jackknife free-response receiver operating characteristics (JAFROC) analysis was conducted to assess observer performance. Compared with interpretation without TS images, interpretation with TS images was associated with a significantly higher mean figure of merit (0.710 vs. 0.658; JAFROC analysis, P = 0.0027). Mean sensitivity at lesion-based was significantly higher for interpretation with TS compared with that without TS (46.1% vs. 33.9%; P = 0.003). Mean false positive count per subject was also significantly higher for interpretation with TS than for that without TS (0.28 vs. 0.15; P

Abstract Endometrial cancer (EC) is the most common gynecological tumor in developed countries, and preoperative risk stratification is essential for personalized medicine. There have been several radiomics studies for noninvasive risk stratification of EC using MRI. Although tumor segmentation is usually necessary for these studies, manual segmentation is not only labor-intensive but may also be subjective. Therefore, our study aimed to perform the automatic segmentation of EC on MRI with a convolutional neural network. The effect of the input image sequence and batch size on the segmentation performance was also investigated. Of 200 patients with EC, 180 patients were used for training the modified U-net model; 20 patients for testing the segmentation performance and the robustness of automatically extracted radiomics features. Using multi-sequence images and larger batch size was effective for improving segmentation accuracy. The mean Dice similarity coefficient, sensitivity, and positive predictive value of our model for the test set were 0.806, 0.816, and 0.834, respectively. The robustness of automatically extracted first-order and shape-based features was high (median ICC = 0.86 and 0.96, respectively). Other high-order features presented moderate-high robustness (median ICC = 0.57–0.93). Our model could automatically segment EC on MRI and extract radiomics features with high reliability.

Purpose: The purpose of this study was to develop and evaluate lung cancer segmentation with a pretrained model and transfer learning. The pretrained model was constructed from an artificial dataset generated using a generative adversarial network (GAN).Materials and Methods: Three public datasets containing images of lung nodules/lung cancers were used: LUNA16 dataset, Decathlon lung dataset, and NSCLC radiogenomics. The LUNA16 dataset was used to generate an artificial dataset for lung cancer segmentation with the help of the GAN and 3D graph cut. Pretrained models were then constructed from the artificial dataset. Subsequently, the main segmentation model was constructed from the pretrained models and the Decathlon lung dataset. Finally, the NSCLC radiogenomics dataset was used to evaluate the main segmentation model. The Dice similarity coefficient (DSC) was used as a metric to evaluate the segmentation performance.Results: The mean DSC for the NSCLC radiogenomics dataset improved overall when using the pretrained models. At maximum, the mean DSC was 0.09 higher with the pretrained model than that without it.Conclusion: The proposed method comprising an artificial dataset and a pretrained model can improve lung cancer segmentation as confirmed in terms of the DSC metric. Moreover, the construction of the artificial dataset for the segmentation using the GAN and 3D graph cut was found to be feasible.

PurposeTo retrospectively assess the repeatability of physiological F-18 labeled fluorodeoxyglucose (FDG) uptake in the skin on positron emission tomography/magnetic resonance imaging (PET/MRI) and explore its regional distribution and relationship with sex and age.MethodsOut of 562 examinations with normal FDG distribution on whole-body PET/MRI, 74 repeated examinations were evaluated to assess the repeatability and regional distribution of physiological skin uptake. Furthermore, 224 examinations were evaluated to compare differences in the uptake due to sex and age. Skin segmentation on PET was performed as body-surface contouring on an MR-based attenuation correction map using an off-line reconstruction software. Bland-Altman plots were created for the repeatability assessment. Kruskal-Wallis test was performed to compare the maximum standardized uptake value (SUVmax) with regional distribution, age, and sex.ResultsThe limits of agreement for the difference in SUVmean and SUVmax of the skin were less than 30%. The highest SUVmax was observed in the face (3.09±1.04), followed by the scalp (2.07±0.53). The SUVmax in the face of boys aged 0-9 years and 10-20 years (1.33±0.64 and 2.05±1.00, respectively) and girls aged 0-9 years (0.98±0.38) was significantly lower than that of men aged ≥20 years and girls aged ≥10 years (pConclusionPET/MRI enabled the quantitative analysis of skin FDG uptake with repeatability. The degree of physiological FDG uptake in the skin was the highest in the face and varied between sexes. Although attention to differences in body habitus between age groups is needed, skin FDG uptake also depended on age.

Abstract Convolutional neural network (CNN), a class of artificial neural networks that has become dominant in various computer vision tasks, is attracting interest across a variety of domains, including radiology. CNN is designed to automatically and adaptively learn spatial hierarchies of features through backpropagation by using multiple building blocks, such as convolution layers, pooling layers, and fully connected layers. This review article offers a perspective on the basic concepts of CNN and its application to various radiological tasks, and discusses its challenges and future directions in the field of radiology. Two challenges in applying CNN to radiological tasks, small dataset and overfitting, will also be covered in this article, as well as techniques to minimize them. Being familiar with the concepts and advantages, as well as limitations, of CNN is essential to leverage its potential in diagnostic radiology, with the goal of augmenting the performance of radiologists and improving patient care. Key Points • Convolutional neural network is a class of deep learning methods which has become dominant in various computer vision tasks and is attracting interest across a variety of domains, including radiology. • Convolutional neural network is composed of multiple building blocks, such as convolution layers, pooling layers, and fully connected layers, and is designed to automatically and adaptively learn spatial hierarchies of features through a backpropagation algorithm. • Familiarity with the concepts and advantages, as well as limitations, of convolutional neural network is essential to leverage its potential to improve radiologist performance and, eventually, patient care.

Many recent studies on medical image processing have involved the use of machine learning (ML) and deep learning (DL) [...]

The purpose of this study was to assess whether homology-based emphysema quantification (HEQ) is significantly associated with lung cancer risk. This retrospective study was approved by our institutional review board. We included 576 patients with lung nodules (317 men and 259 women; age, 66.8 ± 12.3 years), who were selected from a database previously generated for computer-aided diagnosis. Of these, 283 were diagnosed with lung cancer, whereas the remaining 293 showed benign lung nodules. HEQ was performed and percentage of low-attenuation lung area (LAA%) was calculated on the basis of computed tomography scans. Statistical models were constructed to estimate lung cancer risk using logistic regression; sex, age, smoking history (Brinkman index), LAA%, and HEQ were considered independent variables. The following three models were evaluated: the base model (sex, age, and smoking history); the LAA% model (the base model + LAA%); and the HEQ model (the base model + HEQ). Model performance was assessed using receiver operating characteristic analysis and the associated area under the curve (AUC). Differences in AUCs among the models were evaluated using Delong's test. AUCs of the base, LAA%, and HEQ models were 0.585, 0.593, and 0.622, respectively. HEQ coefficient was statistically significant in the HEQ model (P = 0.00487), but LAA% coefficient was not significant in the LAA% model (P = 0.199). Delong's test revealed significant difference in AUCs between the LAA% and HEQ models (P = 0.0455). In conclusion, after adjusting for age, sex, and smoking history (Brinkman index), HEQ was significantly associated with lung cancer risk.

The usefulness of sparse-sampling CT with deep learning-based reconstruction for detection of metastasis of malignant ovarian tumors was evaluated. We obtained contrast-enhanced CT images (n = 141) of ovarian cancers from a public database, whose images were randomly divided into 71 training, 20 validation, and 50 test cases. Sparse-sampling CT images were calculated slice-by-slice by software simulation. Two deep-learning models for deep learning-based reconstruction were evaluated: Residual Encoder-Decoder Convolutional Neural Network (RED-CNN) and deeper U-net. For 50 test cases, we evaluated the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) as quantitative measures. Two radiologists independently performed a qualitative evaluation for the following points: entire CT image quality; visibility of the iliac artery; and visibility of peritoneal dissemination, liver metastasis, and lymph node metastasis. Wilcoxon signed-rank test and McNemar test were used to compare image quality and metastasis detectability between the two models, respectively. The mean PSNR and SSIM performed better with deeper U-net over RED-CNN. For all items of the visual evaluation, deeper U-net scored significantly better than RED-CNN. The metastasis detectability with deeper U-net was more than 95%. Sparse-sampling CT with deep learning-based reconstruction proved useful in detecting metastasis of malignant ovarian tumors and might contribute to reducing overall CT-radiation exposure.

Combinations of data augmentation methods and deep learning architectures for automatic pancreas segmentation on CT images are proposed and evaluated. Images from a public CT dataset of pancreas segmentation were used to evaluate the models. Baseline U-net and deep U-net were chosen for the deep learning models of pancreas segmentation. Methods of data augmentation included conventional methods, mixup, and random image cropping and patching (RICAP). Ten combinations of the deep learning models and the data augmentation methods were evaluated. Four-fold cross validation was performed to train and evaluate these models with data augmentation methods. The dice similarity coefficient (DSC) was calculated between automatic segmentation results and manually annotated labels and these were visually assessed by two radiologists. The performance of the deep U-net was better than that of the baseline U-net with mean DSC of 0.703–0.789 and 0.686–0.748, respectively. In both baseline U-net and deep U-net, the methods with data augmentation performed better than methods with no data augmentation, and mixup and RICAP were more useful than the conventional method. The best mean DSC was obtained using a combination of deep U-net, mixup, and RICAP, and the two radiologists scored the results from this model as good or perfect in 76 and 74 of the 82 cases.

We aimed to evaluate a computer-aided diagnosis (CADx) system for lung nodule classification focussing on (i) usefulness of the conventional CADx system (hand-crafted imaging feature + machine learning algorithm), (ii) comparison between support vector machine (SVM) and gradient tree boosting (XGBoost) as machine learning algorithms, and (iii) effectiveness of parameter optimization using Bayesian optimization and random search. Data on 99 lung nodules (62 lung cancers and 37 benign lung nodules) were included from public databases of CT images. A variant of the local binary pattern was used for calculating a feature vector. SVM or XGBoost was trained using the feature vector and its corresponding label. Tree Parzen Estimator (TPE) was used as Bayesian optimization for parameters of SVM and XGBoost. Random search was done for comparison with TPE. Leave-one-out cross-validation was used for optimizing and evaluating the performance of our CADx system. Performance was evaluated using area under the curve (AUC) of receiver operating characteristic analysis. AUC was calculated 10 times, and its average was obtained. The best averaged AUC of SVM and XGBoost was 0.850 and 0.896, respectively; both were obtained using TPE. XGBoost was generally superior to SVM. Optimal parameters for achieving high AUC were obtained with fewer numbers of trials when using TPE, compared with random search. Bayesian optimization of SVM and XGBoost parameters was more efficient than random search. Based on observer study, AUC values of two board-certified radiologists were 0.898 and 0.822. The results show that diagnostic accuracy of our CADx system was comparable to that of radiologists with respect to classifying lung nodules.

We aimed to describe the development of an inference model for computer-aided diagnosis of lung nodules that could provide valid reasoning for any inferences, thereby improving the interpretability and performance of the system. An automatic construction method was used that considered explanation adequacy and inference accuracy. In addition, we evaluated the usefulness of prior experts' (radiologists') knowledge while constructing the models. In total, 179 patients with lung nodules were included and divided into 79 and 100 cases for training and test data, respectively. F-measure and accuracy were used to assess explanation adequacy and inference accuracy, respectively. For F-measure, reasons were defined as proper subsets of Evidence that had a strong influence on the inference result. The inference models were automatically constructed using the Bayesian network and Markov chain Monte Carlo methods, selecting only those models that met the predefined criteria. During model constructions, we examined the effect of including radiologist's knowledge in the initial Bayesian network models. Performance of the best models in terms of F-measure, accuracy, and evaluation metric were as follows: 0.411, 72.0%, and 0.566, respectively, with prior knowledge, and 0.274, 65.0%, and 0.462, respectively, without prior knowledge. The best models with prior knowledge were then subjectively and independently evaluated by two radiologists using a 5-point scale, with 5, 3, and 1 representing beneficial, appropriate, and detrimental, respectively. The average scores by the two radiologists were 3.97 and 3.76 for the test data, indicating that the proposed computer-aided diagnosis system was acceptable to them. In conclusion, the proposed method incorporating radiologists' knowledge could help in eliminating radiologists' distrust of computer-aided diagnosis and improving its performance.

To utilize Gaussian mixture model (GMM) for the quantification of chronic obstructive pulmonary disease (COPD) and to evaluate the combined use of multiple types of quantification.Eighty-seven patients (67 men, 20 women; age, 67.4 ± 11.0 years) who had undergone computed tomography (CT) and pulmonary function test (PFT) were included. The heterogeneity of CT attenuation in emphysema (HC) was obtained by analyzing a distribution of CT attenuation with GMM. The percentages of low-attenuation volume in the lungs (LAV), wall area of bronchi (WA), and the cross-sectional area of small pulmonary vessels (CSA) were also calculated. The relationships between COPD quantifications and the PFT results were evaluated by Pearson's correlation coefficients and through linear models, with the best models selected using Akaike information criterion (AIC).The correlation coefficients with FEV1 were as follows: LAV, -0.505; HC, -0.277; CSA, 0.384; WA, -0.196. The correlation coefficients with FEV1/FVC were: LAV, -0.640; HC, -0.136; CSA, 0.288; WA, -0.131. For predicting FEV1, the smallest AIC values were obtained in the model with LAV, HC, CSA, and WA. For predicting FEV1/FVC, the smallest AIC values were obtained in the model with LAV and HC. In both models, the coefficient of HC was statistically significant (P-values = 0.000880 and 0.0441 for FEV1 and FEV1/FVC, respectively).GMM was applied to COPD quantification. The results of this study show that COPD severity was associated with HC. In addition, it is shown that the combined use of multiple types of quantification made the evaluation of COPD severity more reliable.

We developed a computer-aided diagnosis (CADx) method for classification between benign nodule, primary lung cancer, and metastatic lung cancer and evaluated the following: (i) the usefulness of the deep convolutional neural network (DCNN) for CADx of the ternary classification, compared with a conventional method (hand-crafted imaging feature plus machine learning), (ii) the effectiveness of transfer learning, and (iii) the effect of image size as the DCNN input. Among 1240 patients of previously-built database, computed tomography images and clinical information of 1236 patients were included. For the conventional method, CADx was performed by using rotation-invariant uniform-pattern local binary pattern on three orthogonal planes with a support vector machine. For the DCNN method, CADx was evaluated using the VGG-16 convolutional neural network with and without transfer learning, and hyperparameter optimization of the DCNN method was performed by random search. The best averaged validation accuracies of CADx were 55.9%, 68.0%, and 62.4% for the conventional method, the DCNN method with transfer learning, and the DCNN method without transfer learning, respectively. For image size of 56, 112, and 224, the best averaged validation accuracy for the DCNN with transfer learning were 60.7%, 64.7%, and 68.0%, respectively. DCNN was better than the conventional method for CADx, and the accuracy of DCNN improved when using transfer learning. Also, we found that larger image sizes as inputs to DCNN improved the accuracy of lung nodule classification.

Objectives: The purpose of this study was to validate a patch-based image denoising method for ultra-low-dose CT images. Neural network with convolutional auto-encoder and pairs of standard-dose CT and ultra-low-dose CT image patches were used for image denoising. The performance of the proposed method was measured by using a chest phantom. Materials and methods: Standard-dose and ultra-low-dose CT images of the chest phantom were acquired. The tube currents for standard-dose and ultra-low-dose CT were 300 and 10 mA, respectively. Ultra-low-dose CT images were denoised with our proposed method using neural network, large-scale nonlocal mean, and block-matching and 3D filtering. Five radiologists and three technologists assessed the denoised ultra-low-dose CT images visually and recorded their subjective impressions of streak artifacts, noise other than streak artifacts, visualization of pulmonary vessels, and overall image quality. Results: For the streak artifacts, noise other than streak artifacts, and visualization of pulmonary vessels, the results of our proposed method were statistically better than those of block-matching and 3D filtering (p-values < 0.05). On the other hand, the difference in the overall image quality between our proposed method and block-matching and 3D filtering was not statistically significant (p-value = 0.07272). The p-values obtained between our proposed method and large-scale nonlocal mean were all less than 0.05. Conclusion: Neural network with convolutional auto-encoder could be trained using pairs of standard-dose and ultra-low-dose CT image patches. According to the visual assessment by radiologists and technologists, the performance of our proposed method was superior to that of large-scale nonlocal mean and block-matching and 3D filtering.

OBJECTIVE:The purpose of this study was to investigate the relationship between visual score of emphysema and homology-based emphysema quantification (HEQ) and evaluate whether visual score was accurately predicted by machine learning and HEQ. MATERIALS AND METHODS:A total of 115 anonymized computed tomography images from 39 patients were obtained from a public database. Emphysema quantification of these images was performed by measuring the percentage of low-attenuation lung area (LAA%). The following values related to HEQ were obtained: nb0 and nb1. LAA% and HEQ were calculated at various threshold levels ranging from -1000 HU to -700 HU. Spearman's correlation coefficients between emphysema quantification and visual score were calculated at the various threshold levels. Visual score was predicted by machine learning and emphysema quantification (LAA% or HEQ). Random Forest was used as a machine learning algorithm, and accuracy of prediction was evaluated by leave-one-patient-out cross validation. The difference in the accuracy was assessed using McNemar's test. RESULTS:The correlation coefficients between emphysema quantification and visual score were as follows: LAA% (-950 HU), 0.567; LAA% (-910 HU), 0.654; LAA% (-875 HU), 0.704; nb0 (-950 HU), 0.552; nb0 (-910 HU), 0.629; nb0 (-875 HU), 0.473; nb1 (-950 HU), 0.149; nb1 (-910 HU), 0.519; and nb1 (-875 HU), 0.716. The accuracy of prediction was as follows: LAA%, 55.7% and HEQ, 66.1%. The difference in accuracy was statistically significant (p = 0.0290). CONCLUSION:LAA% and HEQ at -875 HU showed a stronger correlation with visual score than those at -910 or -950 HU. HEQ was more useful than LAA% for predicting visual score.