External Test Dataset Research Articles

Realization of automatic data cleaning and feedback conditioning for J-TEXT ECEI signals based on machine learning

In recent years, Electron Cyclotron Emission Imaging (ECEI) diagnostics and many other imaging diagnostics have become increasingly important in magnetic confinement fusion research. When the image quality becomes worse due to bad pixels, it becomes an important issue for imaging diagnostics. To automatically identify and classify abnormal ECEI signals, a classification algorithm for ECEI signals based on machine learning was developed for the J-TEXT ECEI diagnostic. Incorporated with the digital control function of J-TEXT ECEI, the channels of low-attenuation saturated signals and weak signals can be corrected by adjusting the attenuation levels. At present, the automatic data cleaning and feedback conditioning unit has been set up and applied to the J-TEXT ECEI. The accuracy rate of the classification algorithm on the external test dataset reaches 93.8%. Feedback conditioning can be completed between two discharge shots. This unit can preprocess the diagnostic data for physical analysis and improve the quality of ECEI signals.

Automated detection and quantification of brain metastases on clinical MRI data using artificial neural networks.

BackgroundReliable detection and precise volumetric quantification of brain metastases (BM) on MRI are essential for guiding treatment decisions. Here we evaluate the potential of artificial neural networks (ANN) for automated detection and quantification of BM.MethodsA consecutive series of 308 patients with BM was used for developing an ANN (with a 4:1 split for training/testing) for automated volumetric assessment of contrast-enhancing tumors (CE) and non-enhancing FLAIR signal abnormality including edema (NEE). An independent consecutive series of 30 patients was used for external testing. Performance was assessed case-wise for CE and NEE and lesion-wise for CE using the case-wise/lesion-wise DICE-coefficient (C/L-DICE), positive predictive value (L-PPV) and sensitivity (C/L-Sensitivity).ResultsThe performance of detecting CE lesions on the validation dataset was not significantly affected when evaluating different volumetric thresholds (0.001–0.2 cm3; P = .2028). The median L-DICE and median C-DICE for CE lesions were 0.78 (IQR = 0.6–0.91) and 0.90 (IQR = 0.85–0.94) in the institutional as well as 0.79 (IQR = 0.67–0.82) and 0.84 (IQR = 0.76–0.89) in the external test dataset. The corresponding median L-Sensitivity and median L-PPV were 0.81 (IQR = 0.63–0.92) and 0.79 (IQR = 0.63–0.93) in the institutional test dataset, as compared to 0.85 (IQR = 0.76–0.94) and 0.76 (IQR = 0.68–0.88) in the external test dataset. The median C-DICE for NEE was 0.96 (IQR = 0.92–0.97) in the institutional test dataset as compared to 0.85 (IQR = 0.72–0.91) in the external test dataset.ConclusionThe developed ANN-based algorithm (publicly available at www.github.com/NeuroAI-HD/HD-BM) allows reliable detection and precise volumetric quantification of CE and NEE compartments in patients with BM.

ATSC-NEX: Automated Time Series Classification With Sequential Model-Based Optimization and Nested Cross-Validation

New methods to perform time series classification arise frequently and multiple state-of-the-art approaches achieve high performance on benchmark datasets with respect to accuracy and computation time. However, often the modeling procedures do not include proper validation but rather rely only on either external test dataset or one-level cross-validation. ATSC-NEX is an automated procedure that employs sequential model-based optimization together with nested cross-validation to build an accurate and properly validated time series classification model. It aims to find an optimal pipeline configuration that includes the selection of input type and settings, as well as model type and hyperparameters. The results of a case study in which a model for the identification of diesel engine type is developed, show that the algorithm can efficiently find a well-performing pipeline configuration. The comparison between ATSC-NEX and some state-of-the-art methods on several benchmark datasets shows that similar accuracy can be achieved.

Identification of Disease-specific Single Amino Acid Polymorphisms Using a Simple Random Forest at Protein-level

Background: The number of human genetic variants deposited into publicly available databases has been increasing exponentially. Among these variants, non-synonymous single nucleotide polymorphisms (nsSNPs), also known as single Amino Acid Polymorphisms (SAPs), have been demonstrated to be strongly correlated with phenotypic variations of traits/diseases. Objective: However, the detailed mechanisms governing the disease association of SAPs remain unclear. Thus, further investigation of new attributes and improvement of the prediction becomes more and more urgent since amount of unknown disease-related SAPs need to be investigated. Methods: Based on the principle of Random Forest (RF), we firstly constructed a new effective prediction model for SAPs associated with a particular disease from protein sequences. Four usual sequence signature extractions were separately performed to select the optimal features. Then SAP peptide lengths from 12 to 202 were also optimized. Results: The optimal models achieve higher than 90% accuracy and Area Under the Curve (AUC) of over 0.9 on all 11 external testing datasets. Finally, the good performance on an independent test set with an accuracy higher than 95% proves the superiority of our method. Conclusion: In this paper, based on Random Forest (RF), we constructed 11 disease-association prediction models for SAPs from the protein sequence level. All models yield prediction accuracy higher than 90% and Area Under the Curve (AUC) more than 0.9. Our method only using the information of protein sequences are more universal than those that depend on some additional information or predictions about the proteins.

Automatic detection of mesiodens on panoramic radiographs using artificial intelligence

This study aimed to develop an artificial intelligence model that can detect mesiodens on panoramic radiographs of various dentition groups. Panoramic radiographs of 612 patients were used for training. A convolutional neural network (CNN) model based on YOLOv3 for detecting mesiodens was developed. The model performance according to three dentition groups (primary, mixed, and permanent dentition) was evaluated, both internally (130 images) and externally (118 images), using a multi-center dataset. To investigate the effect of image preprocessing, contrast-limited histogram equalization (CLAHE) was applied to the original images. The accuracy of the internal test dataset was 96.2% and that of the external test dataset was 89.8% in the original images. For the primary, mixed, and permanent dentition, the accuracy of the internal test dataset was 96.7%, 97.5%, and 93.3%, respectively, and the accuracy of the external test dataset was 86.7%, 95.3%, and 86.7%, respectively. The CLAHE images yielded less accurate results than the original images in both test datasets. The proposed model showed good performance in the internal and external test datasets and had the potential for clinical use to detect mesiodens on panoramic radiographs of all dentition types. The CLAHE preprocessing had a negligible effect on model performance.

Predicting Central Serous Chorioretinopathy Recurrence Using Machine Learning.

Purpose: To predict central serous chorioretinopathy (CSC) recurrence 3 and 6 months after laser treatment by using machine learning.Methods: Clinical and imaging features of 461 patients (480 eyes) with CSC were collected at Zhongshan Ophthalmic Center (ZOC) and Xiamen Eye Center (XEC). The ZOC data (416 eyes of 401 patients) were used as the training dataset and the internal test dataset, while the XEC data (64 eyes of 60 patients) were used as the external test dataset. Six different machine learning algorithms and an ensemble model were trained to predict recurrence in patients with CSC. After completing the initial detailed investigation, we designed a simplified model using only clinical data and OCT features.Results: The ensemble model exhibited the best performance among the six algorithms, with accuracies of 0.941 (internal test dataset) and 0.970 (external test dataset) at 3 months and 0.903 (internal test dataset) and 1.000 (external test dataset) at 6 months. The simplified model showed a comparable level of predictive power.Conclusion: Machine learning achieves high accuracies in predicting the recurrence of CSC patients. The application of an intelligent recurrence prediction model for patients with CSC can potentially facilitate recurrence factor identification and precise individualized interventions.

Abstract 10310: Deep Learning for a Diagnosis of Plaque Erosion in Patients with Acute Coronary Syndromes

Background: The diagnosis of plaque erosion by optical coherence tomography (OCT) in patients with acute coronary syndromes (ACS) requires expertise in image interpretation. Objectives: This study sought to develop a deep learning (DL) algorithm that enables an accurate diagnosis of plaque erosion even for readers without experience in image interpretation. Methods: An image sequence transformer DL algorithm that mirrors the attention given by cardiologists to sequential images during a pull-back, was developed. The DL model was trained and internally validated for a diagnosis of plaque erosion both at the frame-level and at the patient-level in 581 ACS patients from 8 institutions in 4 countries. The model was externally tested in 292 ACS patients from an independent dataset. Training and validation dataset included 237,021 cross-sectional OCT images and testing dataset 65,394 cross-sectional OCT images. The OCT image classification performance was evaluated by area under the receiver-operating characteristic curve (AUC), and sensitivity and specificity. The results were compared against diagnoses made by OCT experts both at the frame-level and at the patient-level. Results: In the training and validation data set, 206 (43.2%) patients had plaque erosion, while in the external testing data set, 86 (29.5%) patients had plaque erosion. In the external testing data set, the DL model diagnosed plaque erosion with an AUC of 0.963 (95% CI, 0.962-0.963), a sensitivity of 89.9% (95% CI, 89.7-90.0), and a specificity of 91.1% (95% CI, 91.1-91.2) at the frame-level. At the patient-level, the DL model detected plaque erosion with an AUC of 0.901 (95% CI, 0.900-0.902), a sensitivity of 89.6% (95% CI, 89.4-89.8), and a specificity of 82.0% (95% CI, 81.9-82.2). Conclusion: A newly developed DL model enabled the accurate diagnosis of plaque erosion. This novel DL model will help cardiologists make an accurate diagnosis of plaque erosion in patients with ACS.

Interinstitutional Portability of a Deep Learning Brain MRI Lesion Segmentation Algorithm.

To assess how well a brain MRI lesion segmentation algorithm trained at one institution performed at another institution, and to assess the effect of multi-institutional training datasets for mitigating performance loss. In this retrospective study, a three-dimensional U-Net for brain MRI abnormality segmentation was trained on data from 293 patients from one institution (IN1) (median age, 54 years; 165 women; patients treated between 2008 and 2018) and tested on data from 51 patients from a second institution (IN2) (median age, 46 years; 27 women; patients treated between 2003 and 2019). The model was then trained on additional data from various sources: (a) 285 multi-institution brain tumor segmentations, (b) 198 IN2 brain tumor segmentations, and (c) 34 IN2 lesion segmentations from various brain pathologic conditions. All trained models were tested on IN1 and external IN2 test datasets, assessing segmentation performance using Dice coefficients. The U-Net accurately segmented brain MRI lesions across various pathologic conditions. Performance was lower when tested at an external institution (median Dice score, 0.70 [IN2] vs 0.76 [IN1]). Addition of 483 training cases of a single pathologic condition, including from IN2, did not raise performance (median Dice score, 0.72; P = .10). Addition of IN2 training data with heterogeneous pathologic features, representing only 10% (34 of 329) of total training data, increased performance to baseline (Dice score, 0.77; P < .001). This final model produced total lesion volumes with a high correlation to the reference standard (Spearman r = 0.98). For brain MRI lesion segmentation, adding a modest amount of relevant training data from an external institution to a previously trained model supported successful application of the model to this external institution.Keywords: Neural Networks, Brain/Brain Stem, Segmentation Supplemental material is available for this article. © RSNA, 2021.

Dynamic stochastic deep learning approaches for predicting geometric changes in head and neck cancer

Objective. Modern radiotherapy stands to benefit from the ability to efficiently adapt plans during treatment in response to setup and geometric variations such as those caused by internal organ deformation or tumor shrinkage. A promising strategy is to develop a framework, which given an initial state defined by patient-attributes, can predict future states based on pre-learned patterns from a well-defined patient population. Approach. Here, we investigate the feasibility of predicting patient anatomical changes, defined as a joint state of volume and daily setup changes, across a fractionated treatment schedule using two approaches. The first is based on a new joint framework employing quantum mechanics in combination with deep recurrent neural networks, denoted QRNN. The second approach is developed based on a classical framework, which models patient changes as a Markov process, denoted MRNN. We evaluated the performance characteristics of these two approaches on a dataset of 125 head and neck cancer patients, which was supplemented by synthetic data generated using a generative adversarial network. Model performance was evaluated using area under the receiver operating characteristic curve (AUC) scores. Main results. The MRNN framework had slightly better performance than the QRNN framework, with MRNN (QRNN) validation AUC scores of 0.742 0.021 (0.675 0.036), 0.709 0.026 (0.656 0.021), 0.724 0.036 (0.652 0.044), and 0.698 0.016 (0.605 0.035) for system state vector sizes of 4, 6, 8, and 10, respectively. Of these, only the results from the two higher order states had statistically significant differences A similar trend was also observed when the models were applied to an external testing dataset of 20 patients, yielding MRNN (QRNN) AUC scores of 0.707 (0.623), 0.687 (0.608), 0.723 (0.669), and 0.697 (0.609) for states vectors sizes of 4, 6, 8, and 10, respectively. Significance. These results suggest that both stochastic models have potential value in predicting patient changes during the course of adaptive radiotherapy.

Multi-task vision transformer using low-level chest X-ray feature corpus for COVID-19 diagnosis and severity quantification.

Developing a robust algorithm to diagnose and quantify the severity of the novel coronavirus disease 2019 (COVID-19) using Chest X-ray (CXR) requires a large number of well-curated COVID-19 datasets, which is difficult to collect under the global COVID-19 pandemic. On the other hand, CXR data with other findings are abundant. This situation is ideally suited for the Vision Transformer (ViT) architecture, where a lot of unlabeled data can be used through structural modeling by the self-attention mechanism. However, the use of existing ViT may not be optimal, as the feature embedding by direct patch flattening or ResNet backbone in the standard ViT is not intended for CXR. To address this problem, here we propose a novel Multi-task ViT that leverages low-level CXR feature corpus obtained from a backbone network that extracts common CXR findings. Specifically, the backbone network is first trained with large public datasets to detect common abnormal findings such as consolidation, opacity, edema, etc. Then, the embedded features from the backbone network are used as corpora for a versatile Transformer model for both the diagnosis and the severity quantification of COVID-19. We evaluate our model on various external test datasets from totally different institutions to evaluate the generalization capability. The experimental results confirm that our model can achieve state-of-the-art performance in both diagnosis and severity quantification tasks with outstanding generalization capability, which are sine qua non of widespread deployment.

Optimal ECG-lead selection increases generalizability of deep learning on ECG abnormality classification.

Deep learning (DL) has achieved promising performance in detecting common abnormalities from the 12-lead electrocardiogram (ECG). However, diagnostic redundancy exists in the 12-lead ECG, which could impose a systematic overfitting on DL, causing poor generalization. We, therefore, hypothesized that finding an optimal lead subset of the 12-lead ECG to eliminate the redundancy would help improve the generalizability of DL-based models. In this study, we developed and evaluated a DL-based model that has a feature extraction stage, an ECG-lead subset selection stage and a decision-making stage to automatically interpret multiple common ECG abnormality types. The data analysed in this study consisted of 6877 12-lead ECG recordings from CPSC 2018 (labelled as normal rhythm or eight types of ECG abnormalities, split into training (approx. 80%), validation (approx. 10%) and test (approx. 10%) sets) and 3998 12-lead ECG recordings from PhysioNet/CinC 2020 (labelled as normal rhythm or four types of ECG abnormalities, used as external text set). The ECG-lead subset selection module was introduced within the proposed model to efficiently constrain model complexity. It detected an optimal 4-lead ECG subset consisting of leads II, aVR, V1 and V4. The proposed model using the optimal 4-lead subset significantly outperformed the model using the complete 12-lead ECG on the validation set and on the external test dataset. The results demonstrated that our proposed model successfully identified an optimal subset of 12-lead ECG; the resulting 4-lead ECG subset improves the generalizability of the DL model in ECG abnormality interpretation. This study provides an outlook on what channels are necessary to keep and which ones may be ignored when considering an automated detection system for cardiac ECG abnormalities. This article is part of the theme issue 'Advanced computation in cardiovascular physiology: new challenges and opportunities'.

Cardiac Radiation Dose and Patient Risk Factors for Predicting Major Adverse Cardiac Events in Lung Cancer Patients

<h3>Purpose/Objective(s)</h3> Thoracic radiotherapy (TRT) is a mainstay treatment for locally advanced non-small cell lung cancer (LA-NSCLC). Mean heart dose (MHD) and left anterior descending (LAD) receiving ≥15 Gy (LADV15) was shown to predict major adverse cardiac events (MACE) in LA-NSCLC patients receiving TRT. We aim to develop a clinically applicable MACE risk prediction score model, including both cardiac dosimetric and baseline risk factors. <h3>Materials/Methods</h3> Retrospective data from 748 consecutive patients with LA-NSCLC treated with curative intent between November 1998 to December 2013 were included. The data was split based on diagnosis dates to allow external validation. Model development was performed on 500 consecutive patients diagnosed before December 2010 using multivariable Cox-proportional hazard model; backward elimination scheme (α = 0.05) was used to select the predictors for the final model. Potential predictors were selected <b>a priori</b> based on prior literature: age, pre-existing coronary heart disease (CHD), Framingham Risk, hypertension (HTN), MHD, LADV15, intensity modulated RT (IMRT) use, and interaction between CHD and LADV15 (CHD:LADV15). Model performance was assessed by the Harrell's c-index and internally validated using leave-one out cross validation (LOOCV). The model was applied to the remaining 248 consecutive patients from the initial cohort as the external validation "test" dataset. <h3>Results</h3> The development and test cohorts had median age 64 vs 66 years (<i>P</i> = 0.02), 51.0 vs. 50.4% females (<i>P</i> = 0.88), and 89.4 vs 89.1% stage III cancer (<i>P</i> = 0.91). The final model incorporated CHD, HTN, LADV15, and CHD:LADV15 (β coefficients: 2.703, 1.129, 0.043, -0.047; all <i>P</i> < 0.001; c-index 0.773). LOOCV Pseudo R² = 0.0195. The c-index on the external test dataset was 0.727. Actuarial 3-year MACE rates were 10.7% for development cohort and 15.4% for test cohort. Stratifying development cohort patients into terciles based on MACE risk prediction scores yielded 3-year MACE rates of 0%, 5.7%, and 25.6% for lowest to highest risk-terciles, respectively; among test cohort, the 3-year MACE rates were 4.9%, 10.5%, and 31.9%. Respectively, the 3-year overall survival rates for lowest to highest MACE risk-terciles were 47.3%, 35.2%, and 32.8% among development cohort, and 52.3%, 43.9% and 30.6% within test cohort (both Log-rank <i>P</i> < 0.05). <h3>Conclusion</h3> The 3-year MACE rates spans 5% to 32% from lowest to highest risk groups. Both LADV15 and pre-existing cardiac risk factors are important in predicting MACE risk post-TRT. MACE risk score was associated with survival. This tool has the potential to estimate personalized LADV15 constraints based on patient risk factors and acceptable MACE risk thresholds (e.g., 5-10%), thus may help identify patients who may benefit most from the application of advanced RT techniques to further reduce LAD dose as a modifiable risk factor during RT planning.

Deep learning evaluation of biomarkers from echocardiogram videos.

BackgroundLaboratory testing is routinely used to assay blood biomarkers to provide information on physiologic state beyond what clinicians can evaluate from interpreting medical imaging. We hypothesized that deep learning interpretation of echocardiogram videos can provide additional value in understanding disease states and can evaluate common biomarkers results.MethodsWe developed EchoNet-Labs, a video-based deep learning algorithm to detect evidence of anemia, elevated B-type natriuretic peptide (BNP), troponin I, and blood urea nitrogen (BUN), as well as values of ten additional lab tests directly from echocardiograms. We included patients (n = 39,460) aged 18 years or older with one or more apical-4-chamber echocardiogram videos (n = 70,066) from Stanford Healthcare for training and internal testing of EchoNet-Lab's performance in estimating the most proximal biomarker result. Without fine-tuning, the performance of EchoNet-Labs was further evaluated on an additional external test dataset (n = 1,301) from Cedars-Sinai Medical Center. We calculated the area under the curve (AUC) of the receiver operating characteristic curve for the internal and external test datasets.FindingsOn the held-out test set of Stanford patients not previously seen during model training, EchoNet-Labs achieved an AUC of 0.80 (0.79-0.81) in detecting anemia (low hemoglobin), 0.86 (0.85-0.88) in detecting elevated BNP, 0.75 (0.73-0.78) in detecting elevated troponin I, and 0.74 (0.72-0.76) in detecting elevated BUN. On the external test dataset from Cedars-Sinai, EchoNet-Labs achieved an AUC of 0.80 (0.77-0.82) in detecting anemia, of 0.82 (0.79-0.84) in detecting elevated BNP, of 0.75 (0.72-0.78) in detecting elevated troponin I, and of 0.69 (0.66-0.71) in detecting elevated BUN. We further demonstrate the utility of the model in detecting abnormalities in 10 additional lab tests. We investigate the features necessary for EchoNet-Labs to make successful detection and identify potential mechanisms for each biomarker using well-known and novel explainability techniques.InterpretationThese results show that deep learning applied to diagnostic imaging can provide additional clinical value and identify phenotypic information beyond current imaging interpretation methods.FundingJ.W.H. and B.H. are supported by the NSF Graduate Research Fellowship. D.O. is supported by NIH K99 HL157421-01. J.Y.Z. is supported by NSF CAREER 1942926, NIH R21 MD012867-01, NIH P30AG059307 and by a Chan-Zuckerberg Biohub Fellowship.

Cancer Risk Estimation Combining Lung Screening CT with Clinical Data Elements.

To develop a model to estimate lung cancer risk using lung cancer screening CT and clinical data elements (CDEs) without manual reading efforts. Two screening cohorts were retrospectively studied: the National Lung Screening Trial (NLST; participants enrolled between August 2002 and April 2004) and the Vanderbilt Lung Screening Program (VLSP; participants enrolled between 2015 and 2018). Fivefold cross-validation using the NLST dataset was used for initial development and assessment of the co-learning model using whole CT scans and CDEs. The VLSP dataset was used for external testing of the developed model. Area under the receiver operating characteristic curve (AUC) and area under the precision-recall curve were used to measure the performance of the model. The developed model was compared with published risk-prediction models that used only CDEs or imaging data alone. The Brock model was also included for comparison by imputing missing values for patients without a dominant pulmonary nodule. A total of 23 505 patients from the NLST (mean age, 62 years ± 5 [standard deviation]; 13 838 men, 9667 women) and 147 patients from the VLSP (mean age, 65 years ± 5; 82 men, 65 women) were included. Using cross-validation on the NLST dataset, the AUC of the proposed co-learning model (AUC, 0.88) was higher than the published models predicted with CDEs only (AUC, 0.69; P < .05) and with images only (AUC, 0.86; P < .05). Additionally, using the external VLSP test dataset, the co-learning model had a higher performance than each of the published individual models (AUC, 0.91 [co-learning] vs 0.59 [CDE-only] and 0.88 [image-only]; P < .05 for both comparisons). The proposed co-learning predictive model combining chest CT images and CDEs had a higher performance for lung cancer risk prediction than models that contained only CDE or only image data; the proposed model also had a higher performance than the Brock model.Keywords: Computer-aided Diagnosis (CAD), CT, Lung, Thorax Supplemental material is available for this article. © RSNA, 2021.

COVID-19 diagnosis from chest x-rays: developing a simple, fast, and accurate neural network

PurposeChest x-rays are a fast and inexpensive test that may potentially diagnose COVID-19, the disease caused by the novel coronavirus. However, chest imaging is not a first-line test for COVID-19 due to low diagnostic accuracy and confounding with other viral pneumonias. Recent research using deep learning may help overcome this issue as convolutional neural networks (CNNs) have demonstrated high accuracy of COVID-19 diagnosis at an early stage.MethodsWe used the COVID-19 Radiography database [36], which contains x-ray images of COVID-19, other viral pneumonia, and normal lungs. We developed a CNN in which we added a dense layer on top of a pre-trained baseline CNN (EfficientNetB0), and we trained, validated, and tested the model on 15,153 X-ray images. We used data augmentation to avoid overfitting and address class imbalance; we used fine-tuning to improve the model’s performance. From the external test dataset, we calculated the model’s accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1-score.ResultsOur model differentiated COVID-19 from normal lungs with 95% accuracy, 90% sensitivity, and 97% specificity; it differentiated COVID-19 from other viral pneumonia and normal lungs with 93% accuracy, 94% sensitivity, and 95% specificity.ConclusionsOur parsimonious CNN shows that it is possible to differentiate COVID-19 from other viral pneumonia and normal lungs on x-ray images with high accuracy. Our method may assist clinicians with making more accurate diagnostic decisions and support chest X-rays as a valuable screening tool for the early, rapid diagnosis of COVID-19.

Digital Gonioscopy Based on Three-dimensional Anterior-Segment OCT: An International Multicenter Study

To develop and evaluate the performance of a 3-dimensional (3D) deep-learning-based automated digital gonioscopy system (DGS) in detecting 2 major characteristics in eyes with suspected primary angle-closure glaucoma (PACG): (1) narrow iridocorneal angles (static gonioscopy, Task I) and (2) peripheral anterior synechiae (PAS) (dynamic gonioscopy, Task II) on OCT scans. International, cross-sectional, multicenter study. A total of 1.112 million images of 8694 volume scans (2294 patients) from 3 centers were included in this study (Task I, training/internal validation/external testing: 4515, 1101, and 2222 volume scans, respectively; Task II, training/internal validation/external testing: 378, 376, and 102 volume scans, respectively). For Task I, a narrow angle was defined as an eye in which the posterior pigmented trabecular meshwork was not visible in more than 180° without indentation in the primary position captured in the dark room from the scans. For Task II, PAS was defined as the adhesion of the iris to the trabecular meshwork. The diagnostic performance of the 3D DGS was evaluated in both tasks with gonioscopic records as reference. The area under the curve (AUC), sensitivity, and specificity of the 3D DGS were calculated. In Task I, 29.4% of patients had a narrow angle. The AUC, sensitivity, and specificity of 3D DGS on the external testing datasets were 0.943 (0.933-0.953), 0.867 (0.838-0.895), and 0.878 (0.859-0.896), respectively. For Task II, 13.8% of patients had PAS. The AUC, sensitivity, and specificity of 3D DGS were 0.902 (0.818-0.985), 0.900 (0.714-1.000), and 0.890 (0.841-0.938), respectively, on the external testing set at quadrant level following normal clinical practice; and 0.885 (0.836-0.933), 0.912 (0.816-1.000), and 0.700 (0.660-0.741), respectively, on the external testing set at clock-hour level. The 3D DGS is effective in detecting eyes with suspected PACG. It has the potential to be used widely in the primary eye care community for screening of subjects at high risk of developing PACG.

Thyroid ultrasound image classification using a convolutional neural network.

BackgroundUltrasound (US) is widely used in the clinical diagnosis of thyroid nodules. Artificial intelligence-powered US is becoming an important issue in the research community. This study aimed to develop an improved deep learning model-based algorithm to classify benign and malignant thyroid nodules (TNs) using thyroid US images.MethodsIn total, 592 patients with 600 TNs were included in the internal training, validation, and testing data set; 187 patients with 200 TNs were recruited for the external test data set. We developed a Visual Geometry Group (VGG)-16T model, based on the VGG-16 architecture, but with additional batch normalization (BN) and dropout layers in addition to the fully connected layers. We conducted a 10-fold cross-validation to analyze the performance of the VGG-16T model using a data set of gray-scale US images from 5 different brands of US machines.ResultsFor the internal data set, the VGG-16T model had 87.43% sensitivity, 85.43% specificity, and 86.43% accuracy. For the external data set, the VGG-16T model achieved an area under the curve (AUC) of 0.829 [95% confidence interval (CI): 0.770–0.879], a radiologist with 15 years’ working experience achieved an AUC of 0.705 (95% CI: 0.659–0.801), a radiologist with 10 years’ experience achieved an AUC of 0.725 (95% CI: 0.653–0.797), and a radiologist with 5 years’ experience achieved an AUC of 0.660 (95% CI: 0.584–0.736).ConclusionsThe VGG-16T model had high specificity, sensitivity, and accuracy in differentiating between malignant and benign TNs. Its diagnostic performance was superior to that of experienced radiologists. Thus, the proposed improved deep-learning model can assist radiologists to diagnose thyroid cancer.

3D multi-scale, multi-task, and multi-label deep learning for prediction of lymph node metastasis in T1 lung adenocarcinoma patients' CT images

The diagnosis of preoperative lymph node (LN) metastasis is crucial to evaluate possible therapy options for T1 lung adenocarcinoma patients. Radiologists preoperatively diagnose LN metastasis by evaluating signs related to LN metastasis, like spiculation or lobulation of pulmonary nodules in CT images. However, this type of evaluation is subjective and time-consuming, which may result in poor consistency and low efficiency of diagnoses. In this study, a 3D Multi-scale, Multi-task, and Multi-label classification network (3M‐CN) was proposed to predict LN metastasis, as well as evaluate multiple related signs of pulmonary nodules in order to improve the accuracy of LN metastasis prediction. The following key approaches were adapted for this method. First, a multi-scale feature fusion module was proposed to aggregate the features from different levels for which different labels be best modeled at different levels; second, an auxiliary segmentation task was applied to force the model to focus more on the nodule region and less on surrounding unrelated structures; and third, a cross-modal integration module called the refine layer was designed to integrate the related risk factors into the model to further improve its confidence level. The 3M‐CN was trained using data from 401 cases and then validated on both internal and external datasets, which consisted of 100 cases and 53 cases, respectively. The proposed 3M‐CN model was then compared with existing state-of-the-art methods for prediction of LN metastasis. The proposed model outperformed other methods, achieving the best performance with AUCs of 0.945 and 0.948 in the internal and external test datasets, respectively. The proposed model not only obtain strong generalization, but greatly enhance the interpretability of the deep learning model, increase doctors' confidence in the model results, conform to doctors' diagnostic process, and may also be transferable to the diagnosis of other diseases.

Deep learning-based endoscopic anatomy classification: an accelerated approach for data preparation and model validation.

Photodocumentation during endoscopy procedures is one of the indicators for endoscopy performance quality; however, this indicator is difficult to measure and audit in the endoscopy unit. Emerging artificial intelligence technology may solve this problem, which requires a large amount of material for model development. We developed a deep learning-based endoscopic anatomy classification system through convolutional neural networks with an accelerated data preparation approach. We retrospectively collected 8,041 images from esophagogastroduodenoscopy (EGD) procedures and labeled them using two experts for nine anatomical locations of the upper gastrointestinal tract. A base model for EGD image multiclass classification was first developed, and an additional 6,091 images were enrolled and classified by the base model. A total of 5,963 images were manually confirmed and added to develop the subsequent enhanced model. Additional internal and external endoscopy image datasets were used to test the model performance. The base model achieved total accuracy of 96.29%. For the enhanced model, the total accuracy was 96.64%. The overall accuracy improved with the enhanced model compared with the base model for the internal test dataset without narrowband images (93.05% vs. 91.25%, p < 0.01) or with narrowband images (92.74% vs. 90.46%, p < 0.01). The total accuracy was 92.56% of the enhanced model on the external test dataset. We constructed a deep learning-based model with an accelerated approach that can be used for quality control in endoscopy units. The model was also validated with both internal and external datasets with high accuracy.

Knee Implant Identification by Fine-Tuning Deep Learning Models.

Identification of implant model from primary knee arthroplasty in pre-op planning of revision surgery is a challenging task with added delay. The direct impact of this inability to identify the implants in time leads to the increase in complexity in surgery. Deep learning in the medical field for diagnosis has shown promising results in getting better with every iteration. This study aims to find an optimal solution for the problem of identification of make and model of knee arthroplasty prosthesis using automated deep learning models. Deep learning algorithms were used to classify knee arthroplasty implant models. The training, validation and test comprised of 1078 radiographs with a total of 6 knee arthroplasty implant models with anterior-posterior (AP) and lateral views. The performance of the model was calculated using accuracy, sensitivity, and area under the receiver-operating characteristic curve (AUC), which were compared against multiple models trained for comparative in-depth analysis with saliency maps for visualization. After training for a total of 30 epochs on all 6 models, the model performing the best obtained an accuracy of 96.38%, the sensitivity of 97.2% and AUC of 0.985 on an external testing dataset consisting of 162 radiographs. The best performing model correctly and uniquely identified the implants which could be visualized using saliency maps. Deep learning models can be used to differentiate between 6 knee arthroplasty implant models. Saliency maps give us a better understanding of which regions the model is focusing on while predicting the results.