Development and multicenter validation of deep convolutional neural network-based detection of colorectal cancer on abdominal CT.

OBJECTIVES: This study aims to develop computer-aided detection (CAD) for colorectal cancer (CRC) using abdominal CT based on a deep convolutional neural network. METHODS: This retrospective study included consecutive patients with colorectal adenocarcinoma who underwent abdominal CT before CRC resection surgery (training set = 379, test set = 103). We customized the 3D U-Net of nnU-Net (CUNET) for CRC detection, which was trained with fivefold cross-validation using annotated CT images. CUNET was validated using datasets covering various clinical situations and institutions: an internal test set (n = 103), internal patients with CRC first determined by CT (n = 54) and asymptomatic CRC (n = 51), and an external validation set from two institutions (n = 60). During each validation, data from the healthy population were added (internal = 60; external = 130). CUNET was compared with other deep CNNs: residual U-Net and EfficientDet. The CAD performances were evaluated using per-CRC sensitivity (true positive/all CRCs), free-response receiver operating characteristic (FROC), and jackknife alternative FROC (JAFROC) curves. RESULTS: CUNET showed a higher maximum per-CRC sensitivity than residual U-Net and EfficientDet (internal test set 91.3% vs. 61.2%, and 64.1%). The per-CRC sensitivity of CUNET at false-positive rates of 3.0 was as follows: internal CRC determined by CT, 89.3%; internal asymptomatic CRC, 87.3%; and external validation, 89.6%. CUNET detected 69.2% (9/13) of CRCs missed by radiologists and 89.7% (252/281) of CRCs from all validation sets. CONCLUSIONS: CUNET can detect CRC on abdominal CT in patients with various clinical situations and from external institutions. KEY POINTS: • Customized 3D U-Net of nnU-Net (CUNET) can be applied to the opportunistic detection of colorectal cancer (CRC) in abdominal CT, helping radiologists detect unexpected CRC. • CUNET showed the best performance at false-positive rates ≥ 3.0, and 30.1% of false-positives were in the colorectum. CUNET detected 69.2% (9/13) of CRCs missed by radiologists and 87.3% (48/55) of asymptomatic CRCs. • CUNET detected CRCs in multiple validation sets composed of varying clinical situations and from different institutions, and CUNET detected 89.7% (252/281) of CRCs from all validation sets.

Deep learning model for differentiating nasal cavity masses based on nasal endoscopy images.

BACKGROUND: Nasal polyps and inverted papillomas often look similar. Clinically, it is difficult to distinguish the masses by endoscopic examination. Therefore, in this study, we aimed to develop a deep learning algorithm for computer-aided diagnosis of nasal endoscopic images, which may provide a more accurate clinical diagnosis before pathologic confirmation of the nasal masses. METHODS: By performing deep learning of nasal endoscope images, we evaluated our computer-aided diagnosis system's assessment ability for nasal polyps and inverted papilloma and the feasibility of their clinical application. We used curriculum learning pre-trained with patches of nasal endoscopic images and full-sized images. The proposed model's performance for classifying nasal polyps, inverted papilloma, and normal tissue was analyzed using five-fold cross-validation. RESULTS: The normal scores for our best-performing network were 0.9520 for recall, 0.7900 for precision, 0.8648 for F1-score, 0.97 for the area under the curve, and 0.8273 for accuracy. For nasal polyps, the best performance was 0.8162, 0.8496, 0.8409, 0.89, and 0.8273, respectively, for recall, precision, F1-score, area under the curve, and accuracy. Finally, for inverted papilloma, the best performance was obtained for recall, precision, F1-score, area under the curve, and accuracy values of 0.5172, 0.8125, 0.6122, 0.83, and 0.8273, respectively. CONCLUSION: Although there were some misclassifications, the results of gradient-weighted class activation mapping were generally consistent with the areas under the curve determined by otolaryngologists. These results suggest that the convolutional neural network is highly reliable in resolving lesion locations in nasal endoscopic images.

Identifying fragile calcifications of the aortic valve in transcatheter aortic valve replacement: iodine concentration of aortic valvular calcification by spectral CT.

OBJECTIVE: To demonstrate the relationship between spectral computed tomography (CT) measured iodine concentration and strength of aortic valvular calcification (AVC) in patients with aortic valve stenosis (AVS). METHODS: A retrospective study was performed on patients who underwent transcatheter aortic valve replacement (TAVR) for symptomatic AVS and underwent both pre and postprocedural electrocardiogram gated CT scans using a spectral CT system. Preprocedural CT was used to evaluate the volume and iodine concentration (IC) in the AVC. Postprocedural CT data were used to calculate the volume reduction percentage (VRP) of AVC. Multiple linear regression analysis was used to identify the independent variables related to the VRP in AVCs. RESULTS: A total of 94 AVCs were selected from 22 patients. The mean volume and IC of the AVCs before TAVR were 0.37 mL ± 0.15 mL and 7 mg/mL ± 10.5 mg/mL, respectively. After TAVR, a median VRP of all 94 AVCs was 18.5%. Multiple linear regression analysis showed that the IC was independently associated with the VRP (coefficient = 1.64, p < 0.001). When an optimal IC cutoff point was set at 4 mg/mL in the assessment of a fragile AVC which showed the VRP was > 18.5%, the sensitivity was 63%; specificity, 91%; positive predictive value, 88%; and negative predictive value, 71%. CONCLUSIONS: When using spectral CT to prepare the TAVR, measuring the IC of the AVC may be useful to assess the probability of AVC deformity after TAVR. KEY POINTS: • A dual-layer detector-based spectral CT enables quantifying iodine of contrast media in the aortic valve calcification (AVC) on contrast-enhanced CT images. • The AVC including iodine of contrast media on contrast-enhanced CT image may have loose compositions, associated with the deformity of AVC after TAVR. • Measuring the iodine concentration in AVC may have the potential to assess the probability of AVC deformity, which may be associated with the outcome and complications after TAVR.2.

Machine Learning-Based Magnetic Resonance Radiomics Analysis for Predicting Low- and High-Grade Clear Cell Renal Cell Carcinoma.

PURPOSE: To explore whether high- and low-grade clear cell renal cell carcinomas (ccRCC) can be distinguished using radiomics features extracted from magnetic resonance imaging. METHODS: In this retrospective study, 154 patients with pathologically proven clear ccRCC underwent contrast-enhanced 3 T magnetic resonance imaging and were assigned to the development (n = 122) and test (n = 32) cohorts in a temporal-split setup. A total of 834 radiomics features were extracted from whole-tumor volumes using 3 sequences: T2-weighted imaging (T2WI), diffusion-weighted imaging, and contrast-enhanced T1-weighted imaging. A random forest regressor was used to extract important radiomics features that were subsequently used for model development using the random forest algorithm. Tumor size, apparent diffusion coefficient value, and percentage of tumor-to-renal parenchymal signal intensity drop in the tumors were recorded by 2 radiologists for quantitative analysis. The area under the receiver operating characteristic curve (AUC) was generated to predict ccRCC grade. RESULTS: In the development cohort, the T2WI-based radiomics model demonstrated the highest performance (AUC, 0.82). The T2WI-based radiomics and radiologic feature hybrid model showed AUCs of 0.79 and 0.83, respectively. In the test cohort, the T2WI-based radiomics model achieved an AUC of 0.82. The range of AUCs of the hybrid model of T2WI-based radiomics and radiologic features was 0.73 to 0.80. CONCLUSION: Magnetic resonance imaging-based classifier models using radiomics features and machine learning showed satisfactory diagnostic performance in distinguishing between high- and low-grade ccRCC, thereby serving as a helpful noninvasive tool for predicting ccRCC grade.

Value of Breast MRI and Nomogram After Negative Axillary Ultrasound for Predicting Axillary Lymph Node Metastasis in Patients With Clinically T1-2 N0 Breast Cancer.

BACKGROUND: There are increasing concerns about that sentinel lymph node biopsy (SLNB) could be omitted in patients with clinically T1-2 N0 breast cancers who has negative axillary ultrasound (AUS). This study aims to assess the false negative result (FNR) of AUS, the rate of high nodal burden (HNB) in clinically T1-2 N0 breast cancer patients, and the diagnostic performance of breast magnetic resonance imaging (MRI) and nomogram. METHODS: We identified 948 consecutive patients with clinically T1-2 N0 cancers who had negative AUS, subsequent MRI, and breast conserving therapy between 2013 and 2020 from two tertiary medical centers. Patients from two centers were assigned to development and validation sets, respectively. Among 948 patients, 402 (mean age ± standard deviation, 57.61 ± 11.58) were within development cohort and 546 (54.43 ± 10.02) within validation cohort. Using logistic regression analyses, clinical-imaging factors associated with lymph node (LN) metastasis were analyzed in the development set from which nomogram was created. The performance of MRI and nomogram was assessed. HNB was defined as ≥ 3 positive LNs. RESULTS: The FNR of AUS was 20.1% (81 of 402) and 19.2% (105 of 546) and the rates of HNB were 1.2% (5/402) and 2.2% (12/546), respectively. Clinical and imaging features associated with LN metastasis were progesterone receptor positivity, outer tumor location on mammography, breast imaging reporting and data system category 5 assessment of cancer on ultrasound, and positive axilla on MRI. In validation cohorts, the positive predictive value (PPV) and negative predictive value (NPV) of MRI and clinical-imaging nomogram was 58.5% and 86.5%, and 56.0% and 82.0%, respectively. CONCLUSION: The FNR of AUS was approximately 20% but the rate of HNB was low. The diagnostic performance of MRI was not satisfactory with low PPV but MRI had merit in reaffirming negative AUS with high NPV. Patients who had low probability scores from our clinical-imaging nomogram might be possible candidates for the omission of SLNB.

Aortic Annulus Detection Based on Deep Learning for Transcatheter Aortic Valve Replacement Using Cardiac Computed Tomography.

BACKGROUND: To propose a deep learning architecture for automatically detecting the complex structure of the aortic annulus plane using cardiac computed tomography (CT) for transcatheter aortic valve replacement (TAVR). METHODS: This study retrospectively reviewed consecutive patients who underwent TAVR between January 2017 and July 2020 at a tertiary medical center. Annulus Detection Permuted AdaIN network (ADPANet) based on a three-dimensional (3D) U-net architecture was developed to detect and localize the aortic annulus plane using cardiac CT. Patients (N = 72) who underwent TAVR between January 2017 and July 2020 at a tertiary medical center were enrolled. Ground truth using a limited dataset was delineated manually by three cardiac radiologists. Training, tuning, and testing sets (70:10:20) were used to build the deep learning model. The performance of ADPANet for detecting the aortic annulus plane was analyzed using the root mean square error (RMSE) and dice similarity coefficient (DSC). RESULTS: In this study, the total dataset consisted of 72 selected scans from patients who underwent TAVR. The RMSE and DSC values for the aortic annulus plane using ADPANet were 55.078 ± 35.794 and 0.496 ± 0.217, respectively. CONCLUSION: Our deep learning framework was feasible to detect the 3D complex structure of the aortic annulus plane using cardiac CT for TAVR. The performance of our algorithms was higher than other convolutional neural networks.

Synthetic Time of Flight Magnetic Resonance Angiography Generation Model Based on Cycle-Consistent Generative Adversarial Network Using PETRA-MRA in the Patients With Treated Intracranial Aneurysm.

BACKGROUND: Pointwise encoding time reduction with radial acquisition (PETRA) magnetic resonance angiography (MRA) is useful for evaluating intracranial aneurysm recurrence, but the problem of severe background noise and low peripheral signal-to-noise ratio (SNR) remain. Deep learning could reduce noise using high- and low-quality images. PURPOSE: To develop a cycle-consistent generative adversarial network (cycleGAN)-based deep learning model to generate synthetic TOF (synTOF) using PETRA. STUDY TYPE: Retrospective. POPULATION: A total of 377 patients (mean age: 60 ± 11; 293 females) with treated intracranial aneurysms who underwent both PETRA and TOF from October 2017 to January 2021. Data were randomly divided into training (49.9%, 188/377) and validation (50.1%, 189/377) groups. FIELD STRENGTH/SEQUENCE: Ultra-short echo time and TOF-MRA on a 3-T MR system. ASSESSMENT: For the cycleGAN model, the peak SNR (PSNR) and structural similarity (SSIM) were evaluated. Image quality was compared qualitatively (5-point Likert scale) and quantitatively (SNR). A multireader diagnostic optimality evaluation was performed with 17 radiologists (experience of 1-18 years). STATISTICAL TESTS: Generalized estimating equation analysis, Friedman's test, McNemar test, and Spearman's rank correlation. P < 0.05 indicated statistical significance. RESULTS: The PSNR and SSIM between synTOF and TOF were 17.51 [16.76; 18.31] dB and 0.71 ± 0.02. The median values of overall image quality, noise, sharpness, and vascular conspicuity were significantly higher for synTOF than for PETRA (4.00 [4.00; 5.00] vs. 4.00 [3.00; 4.00]; 5.00 [4.00; 5.00] vs. 3.00 [2.00; 4.00]; 4.00 [4.00; 4.00] vs. 4.00 [3.00; 4.00]; 3.00 [3.00; 4.00] vs. 3.00 [2.00; 3.00]). The SNRs of the middle cerebral arteries were the highest for synTOF (synTOF vs. TOF vs. PETRA; 63.67 [43.25; 105.00] vs. 52.42 [32.88; 74.67] vs. 21.05 [12.34; 37.88]). In the multireader evaluation, there was no significant difference in diagnostic optimality or preference between synTOF and TOF (19.00 [18.00; 19.00] vs. 20.00 [18.00; 20.00], P = 0.510; 8.00 [6.00; 11.00] vs. 11.00 [9.00, 14.00], P = 1.000). DATA CONCLUSION: The cycleGAN-based deep learning model provided synTOF free from background artifact. The synTOF could be a versatile alternative to TOF in patients who have undergone PETRA for evaluating treated aneurysms. EVIDENCE LEVEL: 4 TECHNICAL EFFICACY: Stage 1.

Machine learning with multiparametric breast MRI for prediction of Ki-67 and histologic grade in early-stage luminal breast cancer.

OBJECTIVES: To investigate whether machine learning-based prediction models using 3-T multiparametric MRI (mpMRI) can predict Ki-67 and histologic grade in stage I-II luminal cancer. METHODS: Between 2013 and 2019, consecutive women with luminal cancers who underwent preoperative MRI with diffusion-weighted imaging (DWI) and surgery were included. For prediction models, morphology, kinetic features using computer-aided diagnosis (CAD), and apparent diffusion coefficient (ADC) at DWI were evaluated by two radiologists. Logistic regression analysis was used to identify mpMRI features for predicting Ki-67 and grade. Diagnostic performance was assessed using eight machine learning algorithms incorporating mpMRI features and compared using the DeLong method. RESULTS: Of 300 women, 203 (67.7%) had low Ki-67 and 97 (32.3%) had high Ki-67; 242 (80.7%) had low grade and 58 (19.3%) had high grade. In multivariate analysis, independent predictors for higher Ki-67 were washout component > 13.5% (odds ratio [OR] = 4.16; p < 0.001) and intratumoral high SI on T2-weighted image (OR = 1.89; p = 0.022). Those for higher grade were washout component > 15.5% (OR = 7.22; p < 0.001), rim enhancement (OR = 2.59; p = 0.022), and ADC value < 0.945 × 10-3 mm2/s (OR = 2.47; p = 0.015). Among eight models using these predictors, six models showed the equivalent performance for Ki-67 (area under the receiver operating characteristic curve [AUC]: 0.70) and Naive Bayes classifier showed the highest performance for grade (AUC: 0.79). CONCLUSIONS: A prediction model incorporating mpMRI features shows good diagnostic performance for predicting Ki-67 and histologic grade in patients with luminal breast cancers. KEY POINTS: • Among multiparametric MRI features, kinetic feature of washout component >13.5% and intratumoral high signal intensity on T2-weighted image were associated with higher Ki-67. • Washout component >15.5%, rim enhancement, and mean apparent diffusion coefficient value < 0.945 × 10-3 mm2/s were associated with higher histologic grade. • Machine learning-based prediction models incorporating multiparametric MRI features showed good diagnostic performance for Ki-67 and histologic grade in luminal breast cancers.

Diagnostic Feasibility of Magnetic Resonance Elastography Radiomics Analysis for the Assessment of Hepatic Fibrosis in Patients With Nonalcoholic Fatty Liver Disease.

OBJECTIVE: The aim of the study was to investigate the diagnostic feasibility of radiomics analysis using magnetic resonance elastography (MRE) to assess hepatic fibrosis in patients with nonalcoholic fatty liver disease (NAFLD). METHODS: One hundred patients with suspected NAFLD were retrospectively enrolled. All patients underwent a liver parenchymal biopsy. Magnetic resonance elastography was performed using a 3.0-T scanner. After multislice segmentation of MRE images, 834 radiomic features were analyzed using a commercial program. Radiologic features, such as median and mean values of the regions of interest and variable clinical features, were analyzed. A random forest regressor was used to extract important radiomic, radiological, and clinical features. A random forest classifier model was trained to use these features to classify the fibrosis stage. The area under the receiver operating characteristic curve was evaluated using a classifier for fibrosis stage diagnosis. RESULTS: The pathological hepatic fibrosis stage was classified as low-grade fibrosis (stages F0-F1, n = 82) or clinically significant fibrosis (stages F2-F4, n = 18). Eight important features were extracted from radiomics analysis, with the 2 most important being wavelet-high high low gray level dependence matrix dependence nonuniformity-normalized and wavelet-high high low gray level dependence matrix dependence entropy. The median value of the multiple small regions of interest was identified as the most important radiologic feature. Platelet count has been identified as an important clinical feature. The area under the receiver operating characteristic curve of the classifier using radiomics was comparable with that of radiologic measures (0.97 ± 0.07 and 0.96 ± 0.06, respectively). CONCLUSIONS: Magnetic resonance elastography radiomics analysis provides diagnostic performance comparable with conventional MRE analysis for the assessment of clinically significant hepatic fibrosis in patients with NAFLD.

Efficient Segmentation for Left Atrium With Convolution Neural Network Based on Active Learning in Late Gadolinium Enhancement Magnetic Resonance Imaging.

BACKGROUND: To propose fully automatic segmentation of left atrium using active learning with limited dataset in late gadolinium enhancement in cardiac magnetic resonance imaging (LGE-CMRI). METHODS: An active learning framework was developed to segment the left atrium in cardiac LGE-CMRI. Patients (n = 98) with atrial fibrillation from the Korea University Anam Hospital were enrolled. First, 20 cases were delineated for ground truths by two experts and used for training a draft model. Second, the 20 cases from the first step and 50 new cases, corrected in a human-in-the-loop manner after predicting using the draft model, were used to train the next model; all 98 cases (70 cases from the second step and 28 new cases) were trained. An additional 20 LGE-CMRI were evaluated in each step. RESULTS: The Dice coefficients for the three steps were 0.85 ± 0.06, 0.89 ± 0.02, and 0.90 ± 0.02, respectively. The biases (95% confidence interval) in the Bland-Altman plots of each step were 6.36% (-14.90-27.61), 6.21% (-9.62-22.03), and 2.68% (-8.57-13.93). Deep active learning-based annotation times were 218 ± 31 seconds, 36.70 ± 18 seconds, and 36.56 ± 15 seconds, respectively. CONCLUSION: Deep active learning reduced annotation time and enabled efficient training on limited LGE-CMRI.

Radiomics Analysis of Magnetic Resonance Proton Density Fat Fraction for the Diagnosis of Hepatic Steatosis in Patients With Suspected Non-Alcoholic Fatty Liver Disease.

BACKGROUND: This study aimed to assess the diagnostic feasibility of radiomics analysis based on magnetic resonance (MR)-proton density fat fraction (PDFF) for grading hepatic steatosis in patients with suspected non-alcoholic fatty liver disease (NAFLD). METHODS: This retrospective study included 106 patients with suspected NAFLD who underwent a hepatic parenchymal biopsy. MR-PDFF and MR spectroscopy were performed on all patients using a 3.0-T scanner. Following whole-volume segmentation of the MR-PDFF images, 833 radiomic features were analyzed using a commercial program. Radiologic features were analyzed, including median and mean values of the multiple regions of interest and variable clinical features. A random forest regressor was used to extract the important radiomic, radiologic, and clinical features. The model was trained using 20 repeated 10-fold cross-validations to classify the NAFLD steatosis grade. The area under the receiver operating characteristic curve (AUROC) was evaluated using a classifier to diagnose steatosis grades. RESULTS: The levels of pathological hepatic steatosis were classified as low-grade steatosis (grade, 0-1; n = 82) and high-grade steatosis (grade, 2-3; n = 24). Fifteen important features were extracted from the radiomic analysis, with the three most important being wavelet-LLL neighboring gray tone difference matrix coarseness, original first-order mean, and 90th percentile. The MR spectroscopy mean value was extracted as a more important feature than the MR-PDFF mean or median in radiologic measures. Alanine aminotransferase has been identified as the most important clinical feature. The AUROC of the classifier using radiomics was comparable to that of radiologic measures (0.94 ± 0.09 and 0.96 ± 0.08, respectively). CONCLUSION: MR-PDFF-derived radiomics may provide a comparable alternative for grading hepatic steatosis in patients with suspected NAFLD.

Active Learning for Efficient Segmentation of Liver with Convolutional Neural Network-Corrected Labeling in Magnetic Resonance Imaging-Derived Proton Density Fat Fraction.

This study aimed to propose an efficient method for self-automated segmentation of the liver using magnetic resonance imaging-derived proton density fat fraction (MRI-PDFF) through deep active learning. We developed an active learning framework for liver segmentation using labeled and unlabeled data in MRI-PDFF. A total of 77 liver samples on MRI-PDFF were obtained from patients with nonalcoholic fatty liver disease. For the training, tuning, and testing of the liver segmentation, the ground truth of 71 (internal) and 6 (external) MRI-PDFF scans for training and testing were verified by an expert reviewer. For 100 randomly selected slices, manual and deep learning (DL) segmentations for visual assessments were classified, ranging from very accurate to mostly accurate. The dice similarity coefficients for each step were 0.69 ± 0.21, 0.85 ± 0.12, and 0.94 ± 0.01, respectively (p-value = 0.1389 between the first step and the second step or p-value = 0.0144 between the first step and the third step for paired t-test), indicating that active learning provides superior performance compared with non-active learning. The biases in the Bland-Altman plots for each step were - 24.22% (from - 82.76 to - 2.70), - 21.29% (from - 59.52 to 3.06), and - 0.67% (from - 10.43 to 4.06). Additionally, there was a fivefold reduction in the required annotation time after the application of active learning (2 min with, and 13 min without, active learning in the first step). The number of very accurate slices for DL (46 slices) was greater than that for manual segmentations (6 slices). Deep active learning enables efficient learning for liver segmentation on a limited MRI-PDFF.

Deep semi-supervised learning for automatic segmentation of inferior alveolar nerve using a convolutional neural network.

BACKGROUND: The inferior alveolar nerve (IAN) innervates and regulates the sensation of the mandibular teeth and lower lip. The position of the IAN should be monitored prior to surgery. Therefore, a study using artificial intelligence (AI) was planned to image and track the position of the IAN automatically for a quicker and safer surgery. METHODS: A total of 138 cone-beam computed tomography datasets (Internal: 98, External: 40) collected from multiple centers (three hospitals) were used in the study. A customized 3D nnU-Net was used for image segmentation. Active learning, which consists of three steps, was carried out in iterations for 83 datasets with cumulative additions after each step. Subsequently, the accuracy of the model for IAN segmentation was evaluated using the 50 datasets. The accuracy by deriving the dice similarity coefficient (DSC) value and the segmentation time for each learning step were compared. In addition, visual scoring was considered to comparatively evaluate the manual and automatic segmentation. RESULTS: After learning, the DSC gradually increased to 0.48 ± 0.11 to 0.50 ± 0.11, and 0.58 ± 0.08. The DSC for the external dataset was 0.49 ± 0.12. The times required for segmentation were 124.8, 143.4, and 86.4 s, showing a large decrease at the final stage. In visual scoring, the accuracy of manual segmentation was found to be higher than that of automatic segmentation. CONCLUSIONS: The deep active learning framework can serve as a fast, accurate, and robust clinical tool for demarcating IAN location.

Fully Automated Lung Lobe Segmentation in Volumetric Chest CT with 3D U-Net: Validation with Intra- and Extra-Datasets.

Lung lobe segmentation in chest CT has been used for the analysis of lung functions and surgical planning. However, accurate lobe segmentation is difficult as 80% of patients have incomplete and/or fake fissures. Furthermore, lung diseases such as chronic obstructive pulmonary disease (COPD) can increase the difficulty of differentiating the lobar fissures. Lobar fissures have similar intensities to those of the vessels and airway wall, which could lead to segmentation error in automated segmentation. In this study, a fully automated lung lobe segmentation method with 3D U-Net was developed and validated with internal and external datasets. The volumetric chest CT scans of 196 normal and mild-to-moderate COPD patients from three centers were obtained. Each scan was segmented using a conventional image processing method and manually corrected by an expert thoracic radiologist to create gold standards. The lobe regions in the CT images were then segmented using a 3D U-Net architecture with a deep convolutional neural network (CNN) using separate training, validation, and test datasets. In addition, 40 independent external CT images were used to evaluate the model. The segmentation results for both the conventional and deep learning methods were compared quantitatively to the gold standards using four accuracy metrics including the Dice similarity coefficient (DSC), Jaccard similarity coefficient (JSC), mean surface distance (MSD), and Hausdorff surface distance (HSD). In internal validation, the segmentation method achieved high accuracy for the DSC, JSC, MSD, and HSD (0.97 ± 0.02, 0.94 ± 0.03, 0.69 ± 0.36, and 17.12 ± 11.07, respectively). In external validation, high accuracy was also obtained for the DSC, JSC, MSD, and HSD (0.96 ± 0.02, 0.92 ± 0.04, 1.31 ± 0.56, and 27.89 ± 7.50, respectively). This method took 6.49 ± 1.19 s and 8.61 ± 1.08 s for lobe segmentation of the left and right lungs, respectively. Although various automatic lung lobe segmentation methods have been developed, it is difficult to develop a robust segmentation method. However, the deep learning-based 3D U-Net method showed reasonable segmentation accuracy and computational time. In addition, this method could be adapted and applied to severe lung diseases in a clinical workflow.

Application of deep learning-based computer-aided detection system: detecting pneumothorax on chest radiograph after biopsy.

OBJECTIVES: To retrospectively evaluate the diagnostic performance of a convolutional neural network (CNN) model in detecting pneumothorax on chest radiographs obtained after percutaneous transthoracic needle biopsy (PTNB) for pulmonary lesions. METHODS: A CNN system for computer-aided diagnosis on chest radiographs was developed using the full 26-layer You Only Look Once model. A total of 1596 chest radiographs with pneumothorax were used for training. To validate the clinical feasibility of this model, follow-up chest radiographs obtained after PTNB for 1333 pulmonary lesions in 1319 patients in 2016 were prepared as an independent test set. Two experienced radiologists determined the presence of pneumothorax by consensus. The diagnostic performance of the CNN model was assessed using the jackknife free-response receiver operating characteristic method. RESULTS: The incidence of pneumothorax was 17.9% (247/1379) on 3-h follow-up chest radiographs and 23.3% (309/1329) on 1-day follow-up chest radiographs. Twenty-three (1.7% of all PTNBs) cases required drainage catheter insertion. Our approach had a sensitivity, a specificity, and an area under the curve (AUC), respectively, of 61.1% (151/247), 93.0% (1053/1132), and 0.898 for 3-h follow-up chest radiographs and 63.4% (196/309), 93.5% (954/1020), and 0.905 for 1-day follow-up chest radiographs. The overall accuracy was 87.3% (1204/1379) for 3-h follow-up radiographs and 86.5% (1150/1329) for 1-day follow-up radiographs. The CNN model found all 23 cases of pneumothorax requiring drainage. CONCLUSIONS: Our CNN model had good performance for detection of pneumothorax on chest radiographs after PTNB, especially for those requiring further procedures. It can be used as a screening tool prior to radiologist interpretation. KEY POINTS: • The CNN model had good performance for detection of pneumothorax on chest radiographs after PTNB and showed high specificity and negative predictive value. • The CNN model found all cases of pneumothorax requiring drainage after PTNB. • The CNN model can be used as a screening tool prior to radiologist interpretation.

Sleep Quality Independently Affects Health-related Quality of Life and Cognitive Function in Korean Female Patients with Rheumatoid Arthritis: a Case-control Study.

BACKGROUND: This study aimed to address sleep quality in patients with rheumatoid arthritis (RA) and to determine how it affects health-related quality of life (HRQoL) and cognitive function. METHODS: One hundred and twenty-three patients with RA and 76 healthy controls were enrolled in this study. Sleep quality was assessed using the Korean version of the Pittsburgh Sleep Quality Index (PSQI). Cognitive function and HRQoL was evaluated by a Korean-Montreal Cognitive Assessment (MoCA-K) and 36-item Short-Form Health Survey (SF-36), respectively. Other clinical, demographic, and laboratory data were obtained from retrospective medical chart review. RESULTS: More patients in the RA group reported poor sleep quality (PSQI > 5) than in the control group (61% [75/123] vs. 39.5% [30/76]; P = 0.003). Total PSQI was also significantly higher in the RA group (median [interquartile range], 7 [5-11] vs. 5 [3-6.75]; P = 0.001). Total PSQI score negatively correlated with MoCA-K score (Spearman's rho (r) = -0.223; P = 0.003) with a physical component summary (PCS) of SF-36 (r = -0.221; P = 0.003) and a mental component summary (MCS) of SF-36 (r = -0.341; P < 0.001), which means that poor sleep quality was associated with poor cognitive function and low HRQoL. CONCLUSION: The findings of this study suggest that poor sleep quality is an independent risk factor for low HRQoL and cognitive dysfunction. Efforts to improve the sleep quality of RA patients seem to be an important aspect of integrative treatment for RA.

The Severity and Pattern of Autonomic Dysfunction in Idiopathic Rapid Eye Movement Sleep Behavior Disorder.

BACKGROUND: Autonomic dysfunction in idiopathic rapid eye movement sleep behavior disorder patients has not yet been quantified. The aim of this study was to characterize dysautonomia in patients with idiopathic rapid eye movement sleep behavior disorder using the Composite Autonomic Severity Score, which is a validated instrument for the quantitation of autonomic failure. METHODS: We prospectively enrolled patients with idiopathic rapid eye movement sleep behavior disorder. A battery of standardized autonomic function tests was performed. Autonomic data obtained from the patients were compared to age- and sex-matched healthy controls. RESULTS: Seventeen patients were enrolled. All but 1 patient showed at least one autonomic deficit. These deficits were predominantly adrenergic and cardiovagal and involved relatively few sudomotor problems. The degree of autonomic dysfunction according to the Composite Autonomic Severity Score was mild to moderate in most patients. CONCLUSIONS: Idiopathic rapid eye movement sleep behavior disorder is linked to mild-to-moderate autonomic dysfunction, which is predominantly adrenergic and cardiovagal.

Durability of Hepatitis B e Antigen Seroconversion in Chronic Hepatitis B Patients Treated with Entecavir or Tenofovir.

INTRODUCTION: Loss of HBeAg and development of anti-HBe (seroconversion) is seen as a milestone and endpoint in the treatment of HBeAg-positive patients with chronic hepatitis B (CHB). Among patients treated with nucleos(t)ide analogs (NA), recurrent viremia is common after discontinuation of therapy. Entecavir (ETV) and tenofovir (TDF) are highly potent NA. The durability of virological response and HBeAg seroconversion in patients treated with these agents is not well studied. METHODS: We retrospectively studied the outcomes of 54 HBeAg-positive CHB patients who were treated with either ETV (n = 30) or TDF (23) or both (n = 1) that achieved virological response and underwent seroconversion and consolidation therapy before cessation of treatment. RESULTS: Only 4 (7%) patients had sustained virological, serological, and biochemical remission. Thirteen patients (24%) continued to have HBV DNA levels below 2000 IU/mL and normal alanine aminotransferase activity (ALT). Thirty-seven patients (69%) developed HBV DNA >2000 IU/mL, with 20 having elevated ALT. Among these 37 patients, 23 (62%) remained HBeAg negative/anti-HBe positive, 12 (32%) became HBeAg positive, and 2 (5%) were HBeAg and anti-HBe negative. Duration of consolidation therapy did not correlate with low versus high level of virological relapse. CONCLUSIONS: Durability of HBeAg seroconversion associated with ETV or TDF was not superior to that reported in patients treated with less potent NA. Our results, aggregated with others, suggest HBeAg seroconversion should not be considered as a treatment endpoint for most HBeAg-positive patients treated with NA. Future updates of treatment guidelines should reconsider HBeAg seroconversion as an endpoint to therapy.

Multimodal integration of neuroimaging and genetic data for the diagnosis of mood disorders based on computer vision models.

Mood disorders, particularly major depressive disorder (MDD) and bipolar disorder (BD), are often underdiagnosed, leading to substantial morbidity. Harnessing the potential of emerging methodologies, we propose a novel multimodal fusion approach that integrates patient-oriented brain structural magnetic resonance imaging (sMRI) scans with DNA whole-exome sequencing (WES) data. Multimodal data fusion aims to improve the detection of mood disorders by employing established deep-learning architectures for computer vision and machine-learning strategies. We analyzed brain imaging genetic data of 321 East Asian individuals, including 147 patients with MDD, 78 patients with BD, and 96 healthy controls. We developed and evaluated six fusion models by leveraging common computer vision models in image classification: Vision Transformer (ViT), Inception-V3, and ResNet50, in conjunction with advanced machine-learning techniques (XGBoost and LightGBM) known for high-dimensional data analysis. Model validation was performed using a 10-fold cross-validation. Our ViT ⊕ XGBoost fusion model with MRI scans, genomic Single Nucleotide polymorphism (SNP) data, and unweighted polygenic risk score (PRS) outperformed baseline models, achieving an incremental area under the curve (AUC) of 0.2162 (32.03% increase) and 0.0675 (+8.19%) and incremental accuracy of 0.1455 (+25.14%) and 0.0849 (+13.28%) compared to SNP-only and image-only baseline models, respectively. Our findings highlight the opportunity to refine mood disorder diagnostics by demonstrating the transformative potential of integrating diverse, yet complementary, data modalities and methodologies.