Improved prediction of clinical pregnancy using artificial intelligence with enhanced inner cell mass and trophectoderm images

This study aimed to assess the performance of an artificial intelligence (AI) model for predicting clinical pregnancy using enhanced inner cell mass (ICM) and trophectoderm (TE) images. In this retrospective study, we included static images of 2555 day-5-blastocysts from seven in vitro fertilization centers in South Korea. The main outcome of the study was the predictive capability of the model to detect clinical pregnancies (gestational sac). Compared with the original embryo images, the use of enhanced ICM and TE images improved the average area under the receiver operating characteristic curve for the AI model from 0.716 to 0.741. Additionally, a gradient-weighted class activation mapping analysis demonstrated that the enhanced image-trained AI model was able to extract features from crucial areas of the embryo in 99% (506/512) of the cases. Particularly, it could extract the ICM and TE. In contrast, the AI model trained on the original images focused on the main areas in only 86% (438/512) of the cases. Our results highlight the potential efficacy of using ICM- and TE-enhanced embryo images when training AI models to predict clinical pregnancy.

Prediction of Cardiovascular Complication in Patients with Newly Diagnosed Type 2 Diabetes Using an XGBoost/GRU-ODE-Bayes-Based Machine-Learning Algorithm

BACKGRUOUND

Cardiovascular disease is life-threatening yet preventable for patients with type 2 diabetes mellitus (T2DM). Because each patient with T2DM has a different risk of developing cardiovascular complications, the accurate stratification of cardiovascular risk is critical. In this study, we proposed cardiovascular risk engines based on machine-learning algorithms for newly diagnosed T2DM patients in Korea.

METHODS

To develop the machine-learning-based cardiovascular disease engines, we retrospectively analyzed 26,166 newly diagnosed T2DM patients who visited Seoul St. Mary's Hospital between July 2009 and April 2019. To accurately measure diabetes-related cardiovascular events, we designed a buffer (1 year), an observation (1 year), and an outcome period (5 years). The entire dataset was split into training and testing sets in an 8:2 ratio, and this procedure was repeated 100 times. The area under the receiver operating characteristic curve (AUROC) was calculated by 10-fold cross-validation on the training dataset.

RESULTS

The machine-learning-based risk engines (AUROC XGBoost=0.781±0.014 and AUROC gated recurrent unit [GRU]-ordinary differential equation [ODE]-Bayes=0.812±0.016) outperformed the conventional regression-based model (AUROC=0.723± 0.036).

CONCLUSION

GRU-ODE-Bayes-based cardiovascular risk engine is highly accurate, easily applicable, and can provide valuable information for the individualized treatment of Korean patients with newly diagnosed T2DM.

Development and Validation of a Convolutional Neural Network Model to Predict a Pathologic Fracture in the Proximal Femur Using Abdomen and Pelvis CT Images of Patients With Advanced Cancer

BACKGROUND

Improvement in survival in patients with advanced cancer is accompanied by an increased probability of bone metastasis and related pathologic fractures (especially in the proximal femur). The few systems proposed and used to diagnose impending fractures owing to metastasis and to ultimately prevent future fractures have practical limitations; thus, novel screening tools are essential. A CT scan of the abdomen and pelvis is a standard modality for staging and follow-up in patients with cancer, and radiologic assessments of the proximal femur are possible with CT-based digitally reconstructed radiographs. Deep-learning models, such as convolutional neural networks (CNNs), may be able to predict pathologic fractures from digitally reconstructed radiographs, but to our knowledge, they have not been tested for this application.

QUESTIONS/PURPOSES

(1) How accurate is a CNN model for predicting a pathologic fracture in a proximal femur with metastasis using digitally reconstructed radiographs of the abdomen and pelvis CT images in patients with advanced cancer? (2) Do CNN models perform better than clinicians with varying backgrounds and experience levels in predicting a pathologic fracture on abdomen and pelvis CT images without any knowledge of the patients' histories, except for metastasis in the proximal femur?

METHODS

A total of 392 patients received radiation treatment of the proximal femur at three hospitals from January 2011 to December 2021. The patients had 2945 CT scans of the abdomen and pelvis for systemic evaluation and follow-up in relation to their primary cancer. In 33% of the CT scans (974), it was impossible to identify whether a pathologic fracture developed within 3 months after each CT image was acquired, and these were excluded. Finally, 1971 cases with a mean age of 59 ± 12 years were included in this study. Pathologic fractures developed within 3 months after CT in 3% (60 of 1971) of cases. A total of 47% (936 of 1971) were women. Sixty cases had an established pathologic fracture within 3 months after each CT scan, and another group of 1911 cases had no established pathologic fracture within 3 months after CT scan. The mean age of the cases in the former and latter groups was 64 ± 11 years and 59 ± 12 years, respectively, and 32% (19 of 60) and 53% (1016 of 1911) of cases, respectively, were female. Digitally reconstructed radiographs were generated with perspective projections of three-dimensional CT volumes onto two-dimensional planes. Then, 1557 images from one hospital were used for a training set. To verify that the deep-learning models could consistently operate even in hospitals with a different medical environment, 414 images from other hospitals were used for external validation. The number of images in the groups with and without a pathologic fracture within 3 months after each CT scan increased from 1911 to 22,932 and from 60 to 720, respectively, using data augmentation methods that are known to be an effective way to boost the performance of deep-learning models. Three CNNs (VGG16, ResNet50, and DenseNet121) were fine-tuned using digitally reconstructed radiographs. For performance measures, the area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, precision, and F1 score were determined. The area under the receiver operating characteristic curve was used to evaluate three CNN models mainly, and the optimal accuracy, sensitivity, and specificity were calculated using the Youden J statistic. Accuracy refers to the proportion of fractures in the groups with and without a pathologic fracture within 3 months after each CT scan that were accurately predicted by the CNN model. Sensitivity and specificity represent the proportion of accurately predicted fractures among those with and without a pathologic fracture within 3 months after each CT scan, respectively. Precision is a measure of how few false-positives the model produces. The F1 score is a harmonic mean of sensitivity and precision, which have a tradeoff relationship. Gradient-weighted class activation mapping images were created to check whether the CNN model correctly focused on potential pathologic fracture regions. The CNN model with the best performance was compared with the performance of clinicians.

RESULTS

DenseNet121 showed the best performance in identifying pathologic fractures; the area under the receiver operating characteristic curve for DenseNet121 was larger than those for VGG16 (0.77 ± 0.07 [95% CI 0.75 to 0.79] versus 0.71 ± 0.08 [95% CI 0.69 to 0.73]; p = 0.001) and ResNet50 (0.77 ± 0.07 [95% CI 0.75 to 0.79] versus 0.72 ± 0.09 [95% CI 0.69 to 0.74]; p = 0.001). Specifically, DenseNet121 scored the highest in sensitivity (0.22 ± 0.07 [95% CI 0.20 to 0.24]), precision (0.72 ± 0.19 [95% CI 0.67 to 0.77]), and F1 score (0.34 ± 0.10 [95% CI 0.31 to 0.37]), and it focused accurately on the region with the expected pathologic fracture. Further, DenseNet121 was less likely than clinicians to mispredict cases in which there was no pathologic fracture than cases in which there was a fracture; the performance of DenseNet121 was better than clinician performance in terms of specificity (0.98 ± 0.01 [95% CI 0.98 to 0.99] versus 0.86 ± 0.09 [95% CI 0.81 to 0.91]; p = 0.01), precision (0.72 ± 0.19 [95% CI 0.67 to 0.77] versus 0.11 ± 0.10 [95% CI 0.05 to 0.17]; p = 0.0001), and F1 score (0.34 ± 0.10 [95% CI 0.31 to 0.37] versus 0.17 ± 0.15 [95% CI 0.08 to 0.26]; p = 0.0001).

CONCLUSION

CNN models may be able to accurately predict impending pathologic fractures from digitally reconstructed radiographs of the abdomen and pelvis CT images that clinicians may not anticipate; this can assist medical, radiation, and orthopaedic oncologists clinically. To achieve better performance, ensemble-learning models using knowledge of the patients' histories should be developed and validated. The code for our model is publicly available online at https://github.com/taehoonko/CNN_path_fx_prediction .

LEVEL OF EVIDENCE

Level III, diagnostic study.

A Photoactivated Protein Degrader for Optical Control of Synaptic Function

Hundreds of proteins determine the function of synapses, and synapses define the neuronal circuits that subserve myriad brain, cognitive, and behavioral functions. It is thus necessary to precisely manipulate specific proteins at specific sub-cellular locations and times to elucidate the roles of particular proteins and synapses in brain function. We developed PHOtochemically TArgeting Chimeras (PHOTACs) as a strategy to optically degrade specific proteins with high spatial and temporal precision. PHOTACs are small molecules that, upon wavelength-selective illumination, catalyze ubiquitylation and degradation of target proteins through endogenous proteasomes. Here we describe the design and chemical properties of a PHOTAC that targets Ca 2+ /calmodulin-dependent protein kinase II alpha (CaMKIIα), which is abundant and crucial for baseline synaptic function of excitatory neurons. We validate the PHOTAC strategy, showing that the CaMKIIα-PHOTAC is effective in mouse brain tissue. Light activation of CaMKIIα-PHOTAC removed CaMKIIα from regions of the mouse hippocampus only within 25 μm of the illuminated brain surface. The optically-controlled degradation decreases synaptic function within minutes of light activation, measured by the light-initiated attenuation of evoked field excitatory postsynaptic potential (fEPSP) responses to physiological stimulation. The PHOTACs methodology should be broadly applicable to other key proteins implicated in synaptic function, especially for evaluating their precise roles in the maintenance of long-term potentiation and memory within subcellular dendritic domains.

Improving Machine Learning Diabetes Prediction Models for the Utmost Clinical Effectiveness

The early prediction of diabetes can facilitate interventions to prevent or delay it. This study proposes a diabetes prediction model based on machine learning (ML) to encourage individuals at risk of diabetes to employ healthy interventions. A total of 38,379 subjects were included. We trained the model on 80% of the subjects and verified its predictive performance on the remaining 20%. Furthermore, the performances of several algorithms were compared, including logistic regression, decision tree, random forest, eXtreme Gradient Boosting (XGBoost), Cox regression, and XGBoost Survival Embedding (XGBSE). The area under the receiver operating characteristic curve (AUROC) of the XGBoost model was the largest, followed by those of the decision tree, logistic regression, and random forest models. For the survival analysis, XGBSE yielded an AUROC exceeding 0.9 for the 2- to 9-year predictions and a C-index of 0.934, while the Cox regression achieved a C-index of 0.921. After lowering the threshold from 0.5 to 0.25, the sensitivity increased from 0.011 to 0.236 for the 2-year prediction model and from 0.607 to 0.994 for the 9-year prediction model, while the specificity showed negligible changes. We developed a high-performance diabetes prediction model that applied the XGBSE algorithm with threshold adjustment. We plan to use this prediction model in real clinical practice for diabetes prevention after simplifying and validating it externally.

Aberrant interaction between TEAD1 and Lamin A/C causes cardiomyopathy

Background

Mutations in the LMNA gene encoding Lamin A/C, a major component of the nuclear lamina, cause laminopathies including dilated cardiomyopathy (DCM). DCM patients with LMNA mutations have particularly severe clinical courses such as heart transplantation and death due to heart failure. However, underlying mechanisms of LMNA-induced DCM remains elusive.

Methods and results

We identified LMNA Q353R mutation in a DCM family with severe heart failure. We generated Q353R heterozygous knock-in mice, which showed sarcomere dysplasia and perinatal lethality. Integrative single-cell analyses of the fetal murine hearts and patient-specific induced pluripotent stem cell-derived cardiomyocytes (iPSCMs) revealed that transcriptional regulation of cardiomyocyte maturation/development genes governed by TEAD1 was impaired in LMNA mutant cardiomyocytes. Protein array and immunostaining uncovered increased binding of TEAD1 to mutant Lamin A/C protein and abnormal localization of TEAD1 at the nuclear periphery. Furthermore, TT-10, a Hippo pathway inhibitor, rescued the dysregulation of cardiac developmental genes in LMNA mutant cardiomyocytes. Single-cell RNA-seq of cardiac tissues from DCM patients with the LMNA Q353R mutation confirmed the dysregulated expression of TEAD1 and its target genes. These results demonstrated abnormal interaction between TEAD1 and mutant Lamin A/C impairs structural maturation of cardiomyocytes and suggests that LMNA Q353R-related DCM can be treated through intervention in the Hippo pathway.

Conclusion

TEAD1 trapped by mutant Lamin A/C protein at the nuclear membrane perturbs transcriptional maturation in LMNA Q353R-related DCM.

Funding Acknowledgement

Type of funding sources: None.

P-233 Deep ensembles-based AI as a tool to support embryo grading and clinical pregnancy prediction

Study question

Can deep learning accurately evaluate embryo grades and predict clinical pregnancy while providing relevant clinical evidence, not just results from a black box?

Summary answer

The sophisticated ensemble method can improve the predictive performance for embryo grades and clinical pregnancy, while providing clinically relevant evidence.

What is known already

Previous studies have shown that AI can predict the IVF outcomes by analyzing the images of embryos. In many literature, AI outperformed human because AI could identify features human eyes could not easily detect. However, clinicians have been cautious to adopt the AI technology due to the black box nature of AI algorithms. In this study, we increased the predictive power of AI as well as providing evidence of the prediction by using deep ensembles and Grad-CAM images.

Study design, size, duration

We performed a retrospective study of single static images of 727 Day 5 blastocysts from 270 patients who underwent single embryo transfer at a single in vitro fertilization (IVF) clinic between January 2015 and March 2021. The images were collected from standard optical light microscopes and matched with metadata such as embryo grades and pregnancy outcomes.

Participants/materials, setting, methods

Two different models were designed: an automatic embryo grading model and a pregnancy prediction model. Embryologists labeled a day 5 embryo “GEM,” a good embryo if 4AA/AB or above in the Gardner system and pregnancy was defined as the presence of a fetal heartbeat (FHB). Deep ensembles were applied by training four convolutional neural networks (CNNs) and Grad-CAM images were extracted from the last layer and reviewed by experts.

Main results and the role of chance

Under several single CNNs, the highest AUROCs of the embryo grading model and the pregnancy prediction model were 0.80 and 0.67, respectively. After applying deep ensembles, the AUROCs of the two models increased to 0.84 and 0.72, respectively. When the F1-score for the positive cases were maximized by adjusting the threshold of ensembles, accuracy, sensitivity and specificity of the embryo grading model were 88.1%, 92.9% and 62.5% respectively. For the pregnancy prediction model, accuracy, sensitivity and specificity were 66.3%, 77.1% and 55.6% respectively. The accuracy of GEM predicting pregnancy for the embryologists and the embryo grading AI model was 47.3% and 59.2%, respectively. It is noteworthy that the AI pregnancy prediction model outperformed the embryologists while successfully auto-grading embryos, a strong evidence that AI considered more features for prediction than what was used for grading. It was also noted from the review of the Grad-CAM images that the both AI models were focusing on the ICM, TE and hatching. Although their area of focus was the same, the pregnancy prediction model was able to make better predictions than the embryologists and the embryo grading model.

Limitations, reasons for caution

This study has limitations as it is a retrospective study performed on embryo images from a single IVF center. In addition, including other variables such as clinical data may enhance the models.

Wider implications of the findings

We showed that deep learning can automatically grade embryos and more accurately predict pregnancy than embryologists. Furthermore, the embryologists confirmed the model was looking at key features like ICM, TE and hatching. Sharing such evidence with clinicians can be a necessary step for AI to be adopted for clinical practice.

Trial registration number

not applicable

Asbestosis diagnosis algorithm combining the lung segmentation method and deep learning model in computed tomography image

BACKGROUND

Early detection of asbestosis is important; hence, quick and accurate diagnostic tools are essential. This study aimed to develop an algorithm that combines lung segmentation and deep learning models that can be utilized as a clinical decision support system (CDSS) for diagnosing patients with asbestosis in segmented computed tomography (CT) images.

METHODS

We accurately segmented the lungs in CT images of patients examined at Seoul St. Mary's Hospital using a threshold-based method. Lungs with asbestosis and normal lungs were classified by applying the segmented image to the long-term recurrent convolutional network deep learning model. Performance was evaluated using the area under the receiver operating characteristic curve (AUROC) and F1 score from the test data.

RESULTS

The algorithm developed using the DenseNet201pre-trained model showed excellent performance, with a sensitivity of 0.962, specificity of 0.975, accuracy of 0.970, AUROC of 0.968, and F1 score of 0.961.

CONCLUSIONS

We developed an algorithm with significantly better diagnostic accuracy than a radiologist (0.970 vs. 0.73-0.79). Our developed algorithm is expected to be an excellent support tool if used as a CDSS to diagnose asbestosis using CT images.

Satisfaction and Usability of an Information and Communications Technology-Based System by Clinically Healthy Patients With COVID-19 and Medical Professionals: Cross-sectional Survey and Focus Group Interview Study

BACKGROUND

Digital health care is an important strategy in the war against COVID-19. South Korea introduced living and treatment support centers (LTSCs) to control regional outbreaks and care for patients with asymptomatic or mild COVID-19. Seoul National University Hospital (SNUH) introduced information and communications technology (ICT)-based solutions to manage clinically healthy patients with COVID-19.

OBJECTIVE

This study aims to investigate satisfaction and usability by patients and health professionals in the optimal use of a mobile app and wearable device that SNUH introduced to the LTSC for clinically healthy patients with COVID-19.

METHODS

Online surveys and focus group interviews were conducted to collect quantitative and qualitative data.

RESULTS

Regarding usability testing of the wearable device, perceived usefulness had the highest mean score of 4.45 (SD 0.57) points out of 5. Regarding usability of the mobile app, perceived usefulness had the highest mean score of 4.62 (SD 0.48) points out of 5. Regarding satisfaction items for the mobile app among medical professionals, the "self-reporting" item had the highest mean score of 4.42 (SD 0.58) points out of 5. In focus group interviews of health care professionals, hospital information system interfacing was the most important functional requirement for ICT-based COVID-19 telemedicine.

CONCLUSIONS

Improvement of patient safety and reduction of the burden on medical staff were the expected positive outcomes. Stability and reliability of the device, patient education, accountability, and reimbursement issues should be considered as part of the development of remote patient monitoring. In responding to a novel contagious disease, telemedicine and a wearable device were shown to be useful during a global crisis.

Markers of Myocardial Damage Predict Mortality in Patients With Aortic Stenosis

BACKGROUND

Cardiovascular magnetic resonance (CMR) is increasingly used for risk stratification in aortic stenosis (AS). However, the relative prognostic power of CMR markers and their respective thresholds remains undefined.

OBJECTIVES

Using machine learning, the study aimed to identify prognostically important CMR markers in AS and their thresholds of mortality.

METHODS

Patients with severe AS undergoing AVR (n = 440, derivation; n = 359, validation cohort) were prospectively enrolled across 13 international sites (median 3.8 years' follow-up). CMR was performed shortly before surgical or transcatheter AVR. A random survival forest model was built using 29 variables (13 CMR) with post-AVR death as the outcome.

RESULTS

There were 52 deaths in the derivation cohort and 51 deaths in the validation cohort. The 4 most predictive CMR markers were extracellular volume fraction, late gadolinium enhancement, indexed left ventricular end-diastolic volume (LVEDVi), and right ventricular ejection fraction. Across the whole cohort and in asymptomatic patients, risk-adjusted predicted mortality increased strongly once extracellular volume fraction exceeded 27%, while late gadolinium enhancement >2% showed persistent high risk. Increased mortality was also observed with both large (LVEDVi >80 mL/m²) and small (LVEDVi ≤55 mL/m²) ventricles, and with high (>80%) and low (≤50%) right ventricular ejection fraction. The predictability was improved when these 4 markers were added to clinical factors (3-year C-index: 0.778 vs 0.739). The prognostic thresholds and risk stratification by CMR variables were reproduced in the validation cohort.

CONCLUSIONS

Machine learning identified myocardial fibrosis and biventricular remodeling markers as the top predictors of survival in AS and highlighted their nonlinear association with mortality. These markers may have potential in optimizing the decision of AVR.

Privacy-Oriented Technique for COVID-19 Contact Tracing (PROTECT) Using Homomorphic Encryption: Design and Development Study

BACKGROUND

Various techniques are used to support contact tracing, which has been shown to be highly effective against the COVID-19 pandemic. To apply the technology, either quarantine authorities should provide the location history of patients with COVID-19, or all users should provide their own location history. This inevitably exposes either the patient's location history or the personal location history of other users. Thus, a privacy issue arises where the public good (via information release) comes in conflict with privacy exposure risks.

OBJECTIVE

The objective of this study is to develop an effective contact tracing system that does not expose the location information of the patient with COVID-19 to other users of the system, or the location information of the users to the quarantine authorities.

METHODS

We propose a new protocol called PRivacy Oriented Technique for Epidemic Contact Tracing (PROTECT) that securely shares location information of patients with users by using the Brakerski/Fan-Vercauteren homomorphic encryption scheme, along with a new, secure proximity computation method.

RESULTS

We developed a mobile app for the end-user and a web service for the quarantine authorities by applying the proposed method, and we verified their effectiveness. The proposed app and web service compute the existence of intersections between the encrypted location history of patients with COVID-19 released by the quarantine authorities and that of the user saved on the user's local device. We also found that this contact tracing smartphone app can identify whether the user has been in contact with such patients within a reasonable time.

CONCLUSIONS

This newly developed method for contact tracing shares location information by using homomorphic encryption, without exposing the location information of patients with COVID-19 and other users. Homomorphic encryption is challenging to apply to practical issues despite its high security value. In this study, however, we have designed a system using the Brakerski/Fan-Vercauteren scheme that is applicable to a reasonable size and developed it to an operable format. The developed app and web service can help contact tracing for not only the COVID-19 pandemic but also other epidemics.

The Experimental Process Design of Artificial Lightweight Aggregates Using an Orthogonal Array Table and Analysis by Machine Learning

The purpose of this study is to experimentally design the drying, calcination, and sintering processes of artificial lightweight aggregates through the orthogonal array, to expand the data using the results, and to model the manufacturing process of lightweight aggregates through machine-learning techniques. The experimental design of the process consisted of L₁₈(3⁶6¹), which means that 3⁶ × 6¹ data can be obtained in 18 experiments using an orthogonal array design. After the experiment, the data were expanded to 486 instances and trained by several machine-learning techniques such as linear regression, random forest, and support vector regression (SVR). We evaluated the predictive performance of machine-learning models by comparing predicted and actual values. As a result, the SVR showed the best performance for predicting measured values. This model also worked well for predictions of untested cases.

Stratifying the prognostic capability of cardiovascular magnetic resonance in severe aortic stenosis: a machine learning approach

Background

Cardiovascular magnetic resonance (CMR) demonstrates promise in improving patient risk stratification in aortic stenosis (AS). We explored whether machine learning might provide further insights into the prognostic capability of CMR parameters.

Methods

Severe AS patients (n=440) undergoing AVR were prospectively enrolled across 10 international sites, and CMR performed prior to AVR. A machine learning prediction model using a random survival forest (RSF) was trained with 29 variables, including 13 CMR, 4 echocardiography, and 12 clinical parameters, using post-AVR mortality as an outcome. The impact of the important variables on the outcome (partial dependency) was examined.

Results

The most predictive CMR parameters in the RSF model were the extracellular volume fraction (ECV%), followed by right ventricular ejection fraction (RVEF), late gadolinium enhancement (LGE%), and indexed left ventricular end-diastolic volume (LVEDVi). Regarding the partial effects, the predicted mortality increased strongly once the ECV% exceeded 26.5% (Figure 1A). The LGE% was associated with an increased risk of mortality, which reached a plateau beyond the level of 2% (Figure 1C). There were U-shaped relationships between mortality and both RVEF and LVEDVi, with the lowest mortality seen at RVEF 70% and LVEDVi 68ml/m2 (Figure 1B, D). These trends of predicted outcomes by each variable were verified in the Kaplan-Meier curves and Cox analyses (Table). In both Cox and RSF models, the predictability was substantially increased when these four CMR parameters were added to conventional clinical risk factors. An AS-CMR risk score comprised of these four parameters presented a stepwise increase in mortality with increasing adverse CMR features (p<0.001).

Conclusions

Our machine learning analysis using RSF has identified ECV%, RVEF, LGE%, and LVEDVi as key prognostic markers in severe AS with a nonlinear influence of each parameter on mortality post-AVR.

Figure 1

Funding Acknowledgement

Type of funding source: Public grant(s) – National budget only. Main funding source(s): This study was supported by grants from the Korean Health Technology R & D Project, Ministry of Health, Welfare & Family Affairs, Republic of Korea (HI16C0225 and HI15C0399) and the National Institute for Health Research (NIHR) infrastructure at Leeds.

Explaining the Rationale of Deep Learning Glaucoma Decisions with Adversarial Examples

PURPOSE

To illustrate what is inside the so-called black box of deep learning models (DLMs) so that clinicians can have greater confidence in the conclusions of artificial intelligence by evaluating adversarial explanation on its ability to explain the rationale of DLM decisions for glaucoma and glaucoma-related findings. Adversarial explanation generates adversarial examples (AEs), or images that have been changed to gain or lose pathologic characteristic-specific traits, to explain the DLM's rationale.

DESIGN

Evaluation of explanation methods for DLMs.

PARTICIPANTS

Health screening participants (n = 1653) at the Seoul National University Hospital Health Promotion Center, Seoul, Republic of Korea.

METHODS

We trained DLMs for referable glaucoma (RG), increased cup-to-disc ratio (ICDR), disc rim narrowing (DRN), and retinal nerve fiber layer defect (RNFLD) using 6430 retinal fundus images. Surveys consisting of explanations using AE and gradient-weighted class activation mapping (GradCAM), a conventional heatmap-based explanation method, were generated for 400 pathologic and healthy patient eyes. For each method, board-trained glaucoma specialists rated location explainability, the ability to pinpoint decision-relevant areas in the image, and rationale explainability, the ability to inform the user on the model's reasoning for the decision based on pathologic features. Scores were compared by paired Wilcoxon signed-rank test.

MAIN OUTCOME MEASURES

Area under the receiver operating characteristic curve (AUC), sensitivities, and specificities of DLMs; visualization of clinical pathologic changes of AEs; and survey scores for locational and rationale explainability.

RESULTS

The AUCs were 0.90, 0.99, 0.95, and 0.79 and sensitivities were 0.79, 1.00, 0.82, and 0.55 at 0.90 specificity for RG, ICDR, DRN, and RNFLD DLMs, respectively. Generated AEs showed valid clinical feature changes, and survey results for location explainability were 3.94 ± 1.33 and 2.55 ± 1.24 using AEs and GradCAMs, respectively, of a possible maximum score of 5 points. The scores for rationale explainability were 3.97 ± 1.31 and 2.10 ± 1.25 for AEs and GradCAM, respectively. Adversarial example provided significantly better explainability than GradCAM.

CONCLUSIONS

Adversarial explanation increased the explainability over GradCAM, a conventional heatmap-based explanation method. Adversarial explanation may help medical professionals understand more clearly the rationale of DLMs when using them for clinical decisions.

Information Technology-Based Management of Clinically Healthy COVID-19 Patients: Lessons From a Living and Treatment Support Center Operated by Seoul National University Hospital

Background

South Korea took preemptive action against coronavirus disease (COVID-19) by implementing extensive testing, thorough epidemiological investigation, strict social distancing, and rapid treatment of patients according to disease severity. The Korean government entrusted large-scale hospitals with the operation of living and treatment support centers (LTSCs) for the management for clinically healthy COVID-19 patients.

Objective

The aim of this paper is to introduce our experience implementing information and communications technology (ICT)-based remote patient management systems at a COVID-19 LTSC.

Methods

We adopted new electronic health record templates, hospital information system (HIS) dashboards, cloud-based medical image sharing, a mobile app, and smart vital sign monitoring devices.

Results

Enhancements were made to the HIS to assist in the workflow and care of patients in the LTSC. A dashboard was created for the medical staff to view the vital signs and symptoms of all patients. Patients used a mobile app to consult with their physician or nurse, answer questionnaires, and input self-measured vital signs; the results were uploaded to the hospital information system in real time. Cloud-based image sharing enabled interoperability between medical institutions. Korea’s strategy of aggressive mitigation has “flattened the curve” of the rate of infection. A multidisciplinary approach was integral to develop systems supporting patient care and management at the living and treatment support center as quickly as possible.

Conclusions

Faced with a novel infectious disease, we describe the implementation and experience of applying an ICT-based patient management system in the LTSC affiliated with Seoul National University Hospital. ICT-based tools and applications are increasingly important in health care, and we hope that our experience will provide insight into future technology-based infectious disease responses.

Transfusion after total knee arthroplasty can be predicted using the machine learning algorithm

PURPOSE

A blood transfusion after total knee arthroplasty (TKA) is associated with an increase in complication and infection rates. However, no studies have been conducted to predict transfusion after TKA using a machine learning algorithm. The purpose of this study was to identify informative preoperative variables to create a machine learning model, and to provide a web-based transfusion risk-assessment system for clinical use.

METHODS

This study retrospectively reviewed 1686 patients who underwent TKA at our institution. Data for 43 preoperative variables, including medication history, laboratory values, and demographic characteristics, were collected. Variable selection was conducted using the recursive feature elimination algorithm. The transfusion group was defined as patients with haemoglobin (Hb) < 7 g/dL after TKA. A predictive model was developed using the gradient boosting machine, and the performance of the model was assessed by the area under the receiver operating characteristic curve (AUC). Data sets from an independent institution were tested with the model for external validation.

RESULTS

Of the 1686 patients who underwent TKA, 108 (6.4%) were categorized into the transfusion group. Six preoperative variables were selected, including preoperative Hb, platelet count, type of surgery, tranexamic acid, age, and body weight. The predictive model demonstrated good predictive performance using the six variables [AUC 0.842; 95% confidence interval (CI) 0.820-0.856]. Performance was also good according to the external validation using 400 data from an independent institution (AUC 0.880; 95% CI 0.844-0.910). This web-based blood transfusion risk-assessment system can be accessed at http://safetka.net.

CONCLUSIONS

A web-based predictive model for transfusion after TKA using a machine learning algorithm was developed using six preoperative variables. The model is simple, has been validated, showed good performance, and can be used before TKA to predict the risk of transfusion and guide appropriate precautions for high-risk patients.

LEVEL OF EVIDENCE

Diagnostic level II.

Unsupervised Cluster Analysis of Patients With Aortic Stenosis Reveals Distinct Population With Different Phenotypes and Outcomes

BACKGROUND

There is a lack of studies investigating the heterogeneity of patients with aortic stenosis (AS). We explored whether cluster analysis identifies distinct subgroups with different prognostic significances in AS.

METHODS

Newly diagnosed patients with moderate or severe AS were prospectively enrolled between 2013 and 2016 (n=398, mean 71 years, 55% male). Among demographics, laboratory, and echocardiography parameters (n=32), 11 variables were selected through dimension reduction and used for unsupervised clustering. Phenotypes and causes of mortality were compared between the clusters.

RESULTS

Three clusters with markedly different features were identified. Cluster 1 (n=60) was predominantly associated with cardiac dysfunction, cluster 2 (n=86) consisted of elderly with comorbidities, especially end-stage renal disease, whereas cluster 3 (n=252) demonstrated neither cardiac dysfunction nor comorbidities. Although AS severity did not differ, there was a significant difference in adverse outcomes between the clusters during a median 2.4 years follow-up (mortality rate, 13.3% versus 19.8% versus 6.0% for cluster 1, 2, and 3, P<0.001). Particularly, compared with cluster 3, cluster 1 was associated with only cardiac mortality (adjusted hazard ratio, 7.37 [95% CI, 2.00-27.13]; P=0.003), whereas cluster 2 was associated with higher noncardiac mortality (adjusted hazard ratio, 3.35 [95% CI, 1.26-8.90]; P=0.015). Phenotypes and association of clusters with specific outcomes were reproduced in an independent validation cohort (n=262).

CONCLUSIONS

Unsupervised cluster analysis of patients with AS revealed 3 distinct groups with different causes of death. This provides a new perspective in the categorization of patients with AS that takes into account comorbidities and extravalvular cardiac dysfunction.

Machine Learning-Based Prediction of Korean Triage and Acuity Scale Level in Emergency Department Patients

Objectives

Triage is a process to accurately assess and classify symptoms to identify and provide rapid treatment to patients. The Korean Triage and Acuity Scale (KTAS) is used as a triage instrument in all emergency centers. The aim of this study was to train and compare machine learning models to predict KTAS levels.

Methods

This was a cross-sectional study using data from a single emergency department of a tertiary university hospital. Information collected during triage was used in the analysis. Logistic regression, random forest, and XGBoost were used to predict the KTAS level.

Results

The models with the highest area under the receiver operating characteristic curve (AUROC) were the random forest and XGBoost models trained on the entire dataset (AUROC = 0.922, 95% confidence interval 0.917-0.925 and AUROC = 0.922, 95% confidence interval 0.918-0.925, respectively). The AUROC of the models trained on the clinical data was higher than that of models trained on text data only, but the models trained on all variables had the highest AUROC among similar machine learning models.

Conclusions

Machine learning can robustly predict the KTAS level at triage, which may have many possibilities for use, and the addition of text data improves the predictive performance compared to that achieved by using structured data alone.

Result and Effectiveness of Malicious E-mail Response Training in a Hospital

Malicious e-mails sent intentionally to institutions have caused an increase in data breaches. Measures against these methods must be taken by healthcare institutions to prevent leakage of sensitive personal medical information. As a form of training, we conducted a phishing simulation to gauge the response of the hospital staff to similar email attacks, and to raise awareness about information security protocols.

Mining affective experience for a kansei design study on a recliner

As the technical performance of products progresses, it is becoming more important to design products that satisfy customers' affective experiences. Hence, many studies about Kansei engineering or Kansei design have been conducted to develop products that can satisfy customers' affective experiences. In the Kansei design method, it is important to select affective variables related to the design elements of the product in order to accurately grasp the emotions of customers. Therefore, this study seeks to develop an affective variable extraction methodology that can reflect users' implicit needs effectively and efficiently. In this study, users' affective variables were extracted from online reviews and classified using a self-organizing map (SOM). For verification, the study selected the Amazon e-commerce service and performed a product experiment on recliners. The experimental results show that the most frequently used affective variable in the use of recliners is 'comfort', which is related to various affective variables. In addition, 15 clusters for affective experiences of recliners extracted from Amazon.com were classified through the SOM. The findings suggest that text mining techniques and the SOM can be used to gather and analyze customers' affective experiences effectively and efficiently. The results of this study can also enhance an understanding of customers' emotions regarding recliners.

Discordance of patient and physician health status concerns in systemic lupus erythematosus

Objectives To compare the health status concerns of patients with systemic lupus erythematosus (SLE) and of their physicians. Methods Cross-sectional questionnaire study of SLE patients and their treating physicians at a tertiary disease-specific outpatient clinic. Patients and physicians completed a questionnaire regarding their concern about specific disease manifestations and impact on quality of life. For each item, degree of concern was rated on a five-point Likert scale and summarized as median (interquartile range). Ratings between patients and physicians were compared using Mann-Whitney U tests. Results A total of 84 patients and 21 physicians participated. Patients' predominant concerns centred on function and fatigue, whereas physicians' concerns focused on SLE-related organ complications. Of the 10 highest ranked patient concerns, only two were common to the 10 highest ranked physician concerns, while physicians rated seven significantly differently; all 10 highest ranked physician concerns were rated significantly lower by patients. The three highest ranked patient concerns (fatigue, pain and feeling worn out) were routinely assessed by 47.6%, 42.9% and 9.5% of physicians, respectively. Conclusion There was significant discordance between SLE patient and physician health status concerns. Items which were ranked highly by patients were not assessed consistently by physicians, highlighting a significant gap in healthcare communication.

Validation of the Lupus Impact Tracker in an Australian patient cohort

OBJECTIVES

The objective of this article is to validate the Lupus Impact Tracker (LIT), a disease-specific patient-reported outcome (PRO) tool, in systemic lupus erythematosus (SLE) patients in a multi-ethnic Australian cohort.

METHODS

Patients attending the Monash Lupus Clinic were asked to complete the LIT, a 10-item PRO. Psychometric testing assessing criterion validity, construct validity, test-retest reliability (TRT) and internal consistency reliability (ICR) were performed. We compared the LIT scores across patient characteristics, and correlations between LIT scores and SLEDAI-2k, PGA, and SLICC-SDI were examined.

RESULTS

LIT data were obtained from 73 patients. Patients were 84% female with a median age of 41 years, and 34% were Asian. The cohort had mild-moderate disease activity with a median (IQR) Systemic Lupus Erythematosus Disease Activity Index-2000 (SLEDAI-2k) of 4 (IQR 2-6). The median LIT score was 32.5 (IQR 17.5-50). LIT demonstrated criterion validity against SLEDAI-2k and SDI. Construct validity assessed by confirmatory factor analysis demonstrated an excellent fit (Goodness of fit index 0.95, Comparative Fit Index 1, Root Mean Square Error of Approximation <0.0001). The LIT demonstrated TRT with an overall intraclass correlation coefficient of 0.986 (95% CI 0.968-0.995). ICR was demonstrated with a Cronbach's alpha of 0.838. Patients with disability, low socioeconomic status, or higher disease activity had significantly worse LIT scores.

CONCLUSION

The LIT demonstrated properties consistent with its being valid in this population. Lower socioeconomic status appears to have a significant impact on patient-reported health-related quality of life in SLE.

Inhibition of prostasin-induced ENaC activities by PN-1 and regulation of PN-1 expression by TGF-beta1 and aldosterone

Prostasin has been shown to regulate sodium handling in the kidney. Recently, a serine protease inhibitor, protease nexin-1 (PN-1), was identified as an endogenous inhibitor for prostasin. Therefore, we hypothesized that PN-1 may regulate sodium reabsorption by reducing prostasin activity, and that expression of PN-1 was regulated by transforming growth factor-beta1 (TGF-beta1) or aldosterone, like prostasin. cRNAs for epithelial sodium channel (ENaC), prostasin, and PN-1 were expressed in Xenopus oocytes, and the amiloride-sensitive sodium currents (I(Na)) were measured. The effect of TGF-beta1 and aldosterone on the mRNA and protein abundance of PN-1 and ENaC was detected by real-time polymerase chain reaction and immunoblotting in M-1 cells. Expression of PN-1 substantially decreased prostasin-induced I(Na) by approximately 68% in oocytes. Treatment of M-1 cells with 20 ng/ml TGF-beta1 significantly increased protein expression of PN-1 by 3.8+/-0.5-fold, whereas administration of 10(-6) M aldosterone markedly decreased protein expression of PN-1 to 53.7+/-6.7%. Basolateral, but not apical, application of TGF-beta1 significantly reduced I(eq). To elucidate the involvement of PN-1 in basal ENaC activity, we silenced the expression of PN-1 by using short-interfering RNA. This increased I(eq) by 1.6+/-0.1-fold. Our study indicates that PN-1 could have a natriuretic role by inhibiting prostasin activity and suggests the possibility that aldosterone and TGF-beta reciprocally regulate the expression of PN-1 in renal epithelial cells contributing to salt retention or natriuresis, respectively by an additional mechanism. PN-1 could represent a new factor that contributes to regulation of ENaC activity in the kidney.

Ultrahigh resolution real time OCT imaging using a compact femtosecond Nd:Glass laser and nonlinear fiber

Ultrahigh resolution, real time OCT imaging is demonstrated using a compact femtosecond Nd:Glass laser that is spectrally broadened in a high numerical aperture single mode fiber. A reflective grating phase delay scanner enables broad bandwidth, high-speed group delay scanning. We demonstrate in vivo, ultrahigh resolution, real time OCT imaging at 1 microm center wavelength with <5 microm axial resolution in free space (<4 microm in tissue). The light source is robust, portable, and well suited for in vivo imaging studies.

The first episode of psychosis: the experience of relatives

OBJECTIVE

The aim was to determine the extent of and the correlates of the distress and impact of care families of patients with first episode psychosis were experiencing when they first came for treatment.

METHOD

Subjects were 238 individuals who had presented with a first episode of psychosis and their family members. Family members were assessed with the Psychological General Well-Being Scale, and the Experience of Caregiving Inventory. Patient data included assessment of positive and negative symptoms, depression, quality of life, and substance use.

RESULTS

Family members of these first-episode patients were experiencing distress and difficulties. It was the family's appraisal of the impact of the illness that was associated with their psychological well-being.

CONCLUSION

As the majority of these first episode families are keen to be involved early and have engaged in an intervention programme, the next step should be an evaluation of their involvement to determine if it is effective.

The effects of polishing and bleaching on the colour of discoloured teeth in vivo

Bleaching is an effective method for restoring the colour of discoloured vital teeth. Power bleaching, in particular, in which a bleaching solution containing 35% hydrogen peroxide is activated by a strong light source using a plasma arc, makes it possible to bleach teeth effectively in a short time. The purpose of this study was to determine how polishing or power bleaching the tooth surface affects tooth colour. The subjects selected were patients who had slightly discoloured teeth. The colour of precisely identified sites on six anterior teeth was measured before treatment, after polishing and after bleaching, to ascertain changes in colour. The measurements revealed that tooth colour changes slightly after polishing, but it shows a much greater change after bleaching, and that the post-bleaching change in tooth colour was caused both by elevation of lightness and reduction of yellowness. They also revealed that the colour difference between pre-treatment and post-bleaching does not depend on the type of tooth. These results suggested that power bleaching is an effective technique for improving slightly discoloured vital teeth, regardless of the type of tooth.

Promotion of breastfeeding intervention trial (PROBIT): a cluster-randomized trial in the Republic of Belarus. Design, follow-up, and data validation

This paper summarizes the objectives, design, follow-up, and data validation of a cluster-randomized trial of a breastfeeding promotion intervention modeled on the WHO/UNICEF Baby-Friendly Hospital Initiative (BFHI). Thirty-four hospitals and their affiliated polyclinics in the Republic of Belarus were randomized to receive BFHI training of medical, midwifery, and nursing staffs (experimental group) or to continue their routine practices (control group). All breastfeeding mother-infant dyads were considered eligible for inclusion in the study if the infant was singleton, born at > or = 37 weeks gestation, weighed > or = 2500 grams at birth, and had a 5-minute Apgar score > or = 5, and neither mother nor infant had a medical condition for which breastfeeding was contraindicated. One experimental and one control site refused to accept their randomized allocation and dropped out of the trial. A total of 17,795 mothers were recruited at the 32 remaining sites, and their infants were followed up at 1, 2, 3, 6, 9, and 12 months of age. To our knowledge, this is the largest randomized trial ever undertaken in area of human milk and lactation. Monitoring visits of all experimental and control maternity hospitals and polyclinics were undertaken prior to recruitment and twice more during recruitment and follow-up to ensure compliance with the randomized allocation. Major study outcomes include the occurrence of > or = 1 episode of gastrointestinal infection, > or = 2 respiratory infections, and the duration of breastfeeding, and are analyzed according to randomized allocation ("intention to treat"). One of the 32 remaining study sites was dropped from the trial because of apparently falsified follow-up data, as suggested by an unrealistically low incidence of infection and unrealistically long duration of breastfeeding, and as confirmed by subsequent data audit of polyclinic charts and interviews with mothers of 64 randomly-selected study infants at the site. Smaller random audits at each of the remaining sites showed extremely high concordance between the PROBIT data forms and both the polyclinic charts and maternal interviews, with no evident difference in under- or over-reporting in experimental vs control sites. Of the 17,046 infants recruited from the 31 participating study sites, 16,491 (96.7%) completed the study and only 555 (3.3%) were lost to follow-up. PROBIT's results should help inform decision-making for clinicians, hospitals, industry, and governments concerning the support, protection, and promotion of breastfeeding.

Autocrine TGFbeta signaling mediates vitamin D3 analog-induced growth inhibition in breast cells

In this study, we address whether TGFbeta signaling mediates vitamin D3 analog-induced growth inhibition in nonmalignant and malignant breast cells. Normal mammary epithelial cells (184), immortalized nonmalignant mammary epithelial cells (184A1 and MCF10A), and breast cancer cells (early passage MCF7: MCF7E) were sensitive to the inhibitory effects of vitamin D3 analogs (EB1089 and MC1288) while late passage MCF7 breast cancer (MCF7L) cells were relatively resistant. A similar pattern of sensitivity to TGFbeta was observed with these cells. Thus, the sensitivity to the vitamin D3 analogs correlated with the sensitivity to TGFbeta. MCF7L TGFbetaRII-transfected cells, which have autocrine TGFbeta activity, were more sensitive to EB1089 than MCF7L cells. TGFbeta neutralizing antibody was found to block the inhibitory effects of these analogs. These results are consistent with the idea that autocrine TGFbeta signaling mediates the anti-proliferative effects of the vitamin D3 analogs in these cells. The expression of TGFbeta isoforms and/or TGFbeta receptors was induced by the analogs in the vitamin D3 and TGFbeta sensitive cells. Vitamin D3 analogs did not induce TGFbeta or TGFbeta receptor expression in the resistant MCF7L cells. Therefore, EB1089 induces autocrine TGFbeta activity through increasing expression of TGFbeta isoforms and/or TGFbeta receptors. In addition, EB1089 induced nuclear VDR protein levels in the sensitive 184A1 cells but not in the resistant MCF7L cells. 184A1 cells were more sensitive to EB1089-induced VDR-dependent transactivation than MCF7L cells as measured by a luciferase reporter construct containing the VDRE, indicating a defect of VDR signaling in MCF7L cells. Smad3, a TGFbeta signaling mediator, coactivated VDR-dependent transactivation in 184A1 cells but not in MCF7L cells. These results indicate that Smad3 coactivates VDR to further enhance TGFbeta signaling and vitamin D3 signaling in the sensitive 184A1 cells. The results also indicate that Smad3 is not of itself sufficient to coactivate VDR in TGFbeta/vitamin D3 resistant MCF7L cells and other factors are required. We found that the PI 3-kinase pathway inhibitor LY29004 inhibited the synergy of TGFbeta and EB1089 on VDR-dependent transactivation activity. This indicates that the crosstalk between TGFbeta and vitamin D signaling is also PI 3-kinase pathway dependent.

Imaging needle for optical coherence tomography

We describe a miniature optical coherence tomography (OCT) imaging needle that can be inserted into solid tissues and organs to permit interstitial imaging of their internal microstructures with micrometer scale resolution and minimal trauma. A novel rotational coupler with a glass capillary tube is also presented that couples light from a rotating single-mode fiber to a stationary one. A prototype needle with a 27-gauge (approximately 410-microm) outer diameter has been developed and is demonstrated for in vivo imaging. The OCT needle can be integrated with standard excisional biopsy devices and used for OCT-guided biopsy.

Intraepidermal expression of basement membrane components in the lesional skin of a patient with dystrophic epidermolysis bullosa

The patient was a 15-year-old male. Since birth, he had developed blistering and erosion of the skin. Biopsy skin specimen of the bullous lesions showed subepidermal blister formation. Electron microscopic examination revealed that tissue separation had occurred at the sublamina densa level. By indirect immunofluorescence using antibodies specific for alpha 6 integrin, laminin 5, type IV collagen, and type VII collagen, all of these basement membrane components were detected as coarse granular intracytoplasmic deposits only in the basal and suprabasal cells of the blister roof. In the non-blistered regions, these basement membrane components showed a linear pattern similar to that seen in normal skin. These findings suggest that intraepidermal expression of basement membrane components was closely related to the blister formation. The biological meaning of intraepidermal expression of basement membrane components were also discussed.

Operational improvement of gynecologic laparoscopic operating room services: an internal review

BACKGROUND

To reorganize reusable laparoscopic instrumentation to promote instrument accessibility, minimize instrument breakage, eliminate infrequently used instruments on permanent trays, and help control maintenance costs.

TECHNIQUE

The Robert Wood Johnson University Hospital Gynecologic Steering Committee evaluated during a 5-month period the storage, use, and maintenance of gynecologic laparoscopic instrument sets used in the surgical suite. Acting on this data, the committee oversaw the following changes. Infrequently used instruments were removed from permanent trays and separately packaged. Two types of gynecologic laparoscopy trays were prepared: one for laparoscopic bilateral tubal ligations and one for both diagnostic and operative laparoscopy. A double-decker compartmentalized tray in which instruments were sterilized and stored replaced the extant single-layer ones in which instruments were stacked on each other. To facilitate instrument identification and function, a surgical manual was compiled with photographs of each instrument and a description of its use.

EXPERIENCE

After implementation of these changes, maintenance and sterilization costs for a 10-month period were compared with those for the previous 10 months. There was a savings of $13,889. The ratio of total costs divided by number of cases performed during the two study periods was also compared. There was a savings of $31 per case.

CONCLUSION

Savings were achieved by reorganizing this operating room's handling of reusable gynecologic laparoscopy equipment. By eliminating infrequently used instruments from the permanent trays and by using a double decker compartmentalized tray that was used during surgery, sterilization, and storage, both sterilization costs and maintenance costs were reduced.