Overcoming the field-of-view to diameter trade-off in microendoscopy via computational optrode-array microscopy.

High-resolution microscopy of deep tissue with large field-of-view (FOV) is critical for elucidating organization of cellular structures in plant biology. Microscopy with an implanted probe offers an effective solution. However, there exists a fundamental trade-off between the FOV and probe diameter arising from aberrations inherent in conventional imaging optics (typically, FOV < 30% of diameter). Here, we demonstrate the use of microfabricated non-imaging probes (optrodes) that when combined with a trained machine-learning algorithm is able to achieve FOV of 1x to 5x the probe diameter. Further increase in FOV is achieved by using multiple optrodes in parallel. With a 1 × 2 optrode array, we demonstrate imaging of fluorescent beads (including 30 FPS video), stained plant stem sections and stained living stems. Our demonstration lays the foundation for fast, high-resolution microscopy with large FOV in deep tissue via microfabricated non-imaging probes and advanced machine learning.

Unassisted Clinicians Versus Deep Learning-Assisted Clinicians in Image-Based Cancer Diagnostics: Systematic Review With Meta-analysis.

BACKGROUND: A number of publications have demonstrated that deep learning (DL) algorithms matched or outperformed clinicians in image-based cancer diagnostics, but these algorithms are frequently considered as opponents rather than partners. Despite the clinicians-in-the-loop DL approach having great potential, no study has systematically quantified the diagnostic accuracy of clinicians with and without the assistance of DL in image-based cancer identification. OBJECTIVE: We systematically quantified the diagnostic accuracy of clinicians with and without the assistance of DL in image-based cancer identification. METHODS: PubMed, Embase, IEEEXplore, and the Cochrane Library were searched for studies published between January 1, 2012, and December 7, 2021. Any type of study design was permitted that focused on comparing unassisted clinicians and DL-assisted clinicians in cancer identification using medical imaging. Studies using medical waveform-data graphics material and those investigating image segmentation rather than classification were excluded. Studies providing binary diagnostic accuracy data and contingency tables were included for further meta-analysis. Two subgroups were defined and analyzed, including cancer type and imaging modality. RESULTS: In total, 9796 studies were identified, of which 48 were deemed eligible for systematic review. Twenty-five of these studies made comparisons between unassisted clinicians and DL-assisted clinicians and provided sufficient data for statistical synthesis. We found a pooled sensitivity of 83% (95% CI 80%-86%) for unassisted clinicians and 88% (95% CI 86%-90%) for DL-assisted clinicians. Pooled specificity was 86% (95% CI 83%-88%) for unassisted clinicians and 88% (95% CI 85%-90%) for DL-assisted clinicians. The pooled sensitivity and specificity values for DL-assisted clinicians were higher than for unassisted clinicians, at ratios of 1.07 (95% CI 1.05-1.09) and 1.03 (95% CI 1.02-1.05), respectively. Similar diagnostic performance by DL-assisted clinicians was also observed across the predefined subgroups. CONCLUSIONS: The diagnostic performance of DL-assisted clinicians appears better than unassisted clinicians in image-based cancer identification. However, caution should be exercised, because the evidence provided in the reviewed studies does not cover all the minutiae involved in real-world clinical practice. Combining qualitative insights from clinical practice with data-science approaches may improve DL-assisted practice, although further research is required. TRIAL REGISTRATION: PROSPERO CRD42021281372; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=281372.

Development and validation of an RBP gene signature for prognosis prediction in colorectal cancer based on WGCNA.

BACKGROUND: RNA binding proteins (RBPs) have been implicated in oncogenesis and progression in various cancers. However, the potential value of RBPs as prognostic indicators and therapeutic targets in colorectal cancer (CRC) requires further investigation. METHODS: Four thousand eighty two RBPs were collected from literature. The weighted gene co-expression network analysis (WGCNA) was performed to identify prognosis-related RBP gene modules based on the data attained from the TCGA cohorts. LASSO algorithm was conducted to establish a prognostic risk model, and the validity of the proposed model was confirmed by an independent GEO dataset. Functional enrichment analysis was performed to reveal the potential biological functions and pathways of the signature and to estimate tumor immune infiltration. Potential therapeutic compounds were inferred utilizing CMap database. Expressions of hub genes were further verified through the Human Protein Atlas (HPA) database and RT-qPCR. RESULTS: One thousand seven hundred thirty four RBPs were differently expressed in CRC samples and 4 gene modules remarkably linked to the prognosis were identified, based on which a 12-gene signature was established for prognosis prediction. Multivariate Cox analysis suggested this signature was an independent predicting factor of overall survival (P < 0.001; HR:3.682; CI:2.377-5.705) and ROC curves indicated it has an effective predictive performance (1-year AUC: 0.653; 3-year AUC:0.673; 5-year AUC: 0.777). GSEA indicated that high risk score was correlated with several cancer-related pathways, including cytokine-cytokine receptor cross talk, ECM receptor cross talk, HEDGEHOG signaling cascade and JAK/STAT signaling cascade. ssGSEA analysis exhibited a significant correlation between immune status and the risk signature. Noscapine and clofazimine were screened as potential drugs for CRC patients with high-risk scores. TDRD5 and GPC1 were identified as hub genes and their expression were validated in 15 pairs of surgically resected CRC tissues. CONCLUSION: Our research provides a depth insight of RBPs' role in CRC and the proposed signature are helpful to the personalized treatment and prognostic judgement.

Ultrasound carotid plaque segmentation via image reconstruction-based self-supervised learning with limited training labels.

Carotid total plaque area (TPA) is an important contributing measurement to the evaluation of stroke risk. Deep learning provides an efficient method for ultrasound carotid plaque segmentation and TPA quantification. However, high performance of deep learning requires datasets with many labeled images for training, which is very labor-intensive. Thus, we propose an image reconstruction-based self-supervised learning algorithm (IR-SSL) for carotid plaque segmentation when few labeled images are available. IR-SSL consists of pre-trained and downstream segmentation tasks. The pre-trained task learns region-wise representations with local consistency by reconstructing plaque images from randomly partitioned and disordered images. The pre-trained model is then transferred to the segmentation network as the initial parameters in the downstream task. IR-SSL was implemented with two networks, UNet++ and U-Net, and evaluated on two independent datasets of 510 carotid ultrasound images from 144 subjects at SPARC (London, Canada) and 638 images from 479 subjects at Zhongnan hospital (Wuhan, China). Compared to the baseline networks, IR-SSL improved the segmentation performance when trained on few labeled images (n = 10, 30, 50 and 100 subjects). For 44 SPARC subjects, IR-SSL yielded Dice-similarity-coefficients (DSC) of 80.14-88.84%, and algorithm TPAs were strongly correlated (r=0.962-0.993, p < 0.001) with manual results. The models trained on the SPARC images but applied to the Zhongnan dataset without retraining achieved DSCs of 80.61-88.18% and strong correlation with manual segmentation (r=0.852-0.978, p < 0.001). These results suggest that IR-SSL could improve deep learning when trained on small labeled datasets, making it useful for monitoring carotid plaque progression/regression in clinical use and trials.

An improved MOPSO approach with adaptive strategy for identifying biomarkers from gene expression dataset.

Biomarkers plays an important role in the prediction and diagnosis of cancers. Therefore, it is urgent to design effective methods to extract biomarkers. The corresponding pathway information of the microarray gene expression data can be obtained from public database, which makes possible to identify biomarkers based on pathway information and has been attracted extensive attention. In the most existing methods, all the member genes in the same pathway are regarded as equally important for inferring pathway activity. However, the contribution of each gene should be different in the process of inferring pathway activity. In this research, an improved multi-objective particle swarm optimization algorithm with penalty boundary intersection decomposition mechanism (IMOPSO-PBI) has been proposed to quantify the relevance of each gene in pathway activity inference. In the proposed algorithm, two optimization objectives namely t-score and z-score respectively has been introduced. In addition, in order to solve the problem that optimal set with poor diversity in the most multi-objective optimization algorithms, an adaptive mechanism for adjusting penalty parameters based on PBI decomposition has been introduced. The performance of the proposed IMOPSO-PBI approach compared with some existing methods on six gene expression datasets has been given. To verify the effectiveness of the proposed IMOPSO-PBI algorithm, experiments were carried out on six gene datasets and the results has been compared with the existing methods. The comparative experiment results show that the proposed IMOPSO-PBI method has a higher classification accuracy and the extracted feature genes are verified possess biological significance.

PrivacyMask: Real-world privacy protection in face ID systems.

Recent works have illustrated that many facial privacy protection methods are effective in specific face recognition algorithms. However, the COVID-19 pandemic has promoted the rapid innovation of face recognition algorithms for face occlusion, especially for the face wearing a mask. It is tricky to avoid being tracked by artificial intelligence only through ordinary props because many facial feature extractors can determine the ID only through a tiny local feature. Therefore, the ubiquitous high-precision camera makes privacy protection worrying. In this paper, we establish an attack method directed against liveness detection. A mask printed with a textured pattern is proposed, which can resist the face extractor optimized for face occlusion. We focus on studying the attack efficiency in adversarial patches mapping from two-dimensional to three-dimensional space. Specifically, we investigate a projection network for the mask structure. It can convert the patches to fit perfectly on the mask. Even if it is deformed, rotated and the lighting changes, it will reduce the recognition ability of the face extractor. The experimental results show that the proposed method can integrate multiple types of face recognition algorithms without significantly reducing the training performance. If we combine it with the static protection method, people can prevent face data from being collected.

Lesion detection of chest X-Ray based on scalable attention residual CNN.

Most of the research on disease recognition in chest X-rays is limited to segmentation and classification, but the problem of inaccurate recognition in edges and small parts makes doctors spend more time making judgments. In this paper, we propose a lesion detection method based on a scalable attention residual CNN (SAR-CNN), which uses target detection to identify and locate diseases in chest X-rays and greatly improves work efficiency. We designed a multi-convolution feature fusion block (MFFB), tree-structured aggregation module (TSAM), and scalable channel and spatial attention (SCSA), which can effectively alleviate the difficulties in chest X-ray recognition caused by single resolution, weak communication of features of different layers, and lack of attention fusion, respectively. These three modules are embeddable and can be easily combined with other networks. Through a large number of experiments on the largest public lung chest radiograph detection dataset, VinDr-CXR, the mean average precision (mAP) of the proposed method was improved from 12.83% to 15.75% in the case of the PASCAL VOC 2010 standard, with IoU > 0.4, which exceeds the existing mainstream deep learning model. In addition, the proposed model has a lower complexity and faster reasoning speed, which is conducive to the implementation of computer-aided systems and provides referential solutions for relevant communities.

Research on medical data security sharing scheme based on homomorphic encryption.

With the deep integration of "AI + medicine", AI-assisted technology has been of great help to human beings in the medical field, especially in the area of predicting and diagnosing diseases based on big data, because it is faster and more accurate. However, concerns about data security seriously hinder data sharing among medical institutions. To fully exploit the value of medical data and realize data collaborative sharing, we developed a medical data security sharing scheme based on the C/S communication mode and constructed a federated learning architecture that uses homomorphic encryption technology to protect training parameters. Here, we chose the Paillier algorithm to realize the additive homomorphism to protect the training parameters. Clients do not need to share local data, but only upload the trained model parameters to the server. In the process of training, a distributed parameter update mechanism is introduced. The server is mainly responsible for issuing training commands and weights, aggregating the local model parameters from the clients and predicting the joint diagnostic results. The client mainly uses the stochastic gradient descent algorithm for gradient trimming, updating and transmitting the trained model parameters back to the server. In order to test the performance of this scheme, a series of experiments was conducted. From the simulation results, we can know that the model prediction accuracy is related to the global training rounds, learning rate, batch size, privacy budget parameters etc. The results show that this scheme realizes data sharing while protecting data privacy, completes the accurate prediction of diseases and has a good performance.

Cervical cell extraction network based on optimized yolo.

Early screening for cervical cancer is a common form of cancer prevention. In the microscopic images of cervical cells, the number of abnormal cells is small, and some abnormal cells are heavily stacked. How to solve the segmentation of highly overlapping cells and realize the identification of single cells from overlapping cells is still a heavy task. Therefore, this paper proposes an object detection algorithm of Cell_yolo to effectively and accurately segment overlapping cells. Cell_yolo adopts a simplified network structure and improves the maximum pooling operation, so that the information of the image is preserved to the greatest extent during the model pooling process. Aiming at the characteristics of many overlapping cells in cervical cell images, a non-maximum suppression method of center distance is proposed to prevent the overlapping cell detection frame from being deleted by mistake. At the same time, the loss function is improved and the focus loss function is added to alleviate the imbalance of positive and negative samples in the training process. Experiments are conducted on a private dataset (BJTUCELL). Experiments have verified that the Cell_yolo model has the advantages of low computational complexity and high detection accuracy, and it is superior to common network models such as YOLOv4 and Faster_RCNN.

A weakly supervised learning-based segmentation network for dental diseases.

With the development of deep learning, medical image segmentation has become a promising technique for computer-aided medical diagnosis. However, the supervised training of the algorithm relies on a large amount of labeled data, and the private dataset bias generally exists in previous research, which seriously affects the algorithm's performance. In order to alleviate this problem and improve the robustness and generalization of the model, this paper proposes an end-to-end weakly supervised semantic segmentation network to learn and infer mappings. Firstly, an attention compensation mechanism (ACM) aggregating the class activation map (CAM) is designed to learn complementarily. Then the conditional random field (CRF) is introduced to prune the foreground and background regions. Finally, the obtained high-confidence regions are used as pseudo labels for the segmentation branch to train and optimize using a joint loss function. Our model achieves a Mean Intersection over Union (MIoU) score of 62.84% in the segmentation task, which is an effective improvement of 11.18% compared to the previous network for segmenting dental diseases. Moreover, we further verify that our model has higher robustness to dataset bias by improved localization mechanism (CAM). The research shows that our proposed approach improves the accuracy and robustness of dental disease identification.

A multi-feature fusion decoding study for unilateral upper-limb fine motor imagery.

To address the fact that the classical motor imagination paradigm has no noticeable effect on the rehabilitation training of upper limbs in patients after stroke and the corresponding feature extraction algorithm is limited to a single domain, this paper describes the design of a unilateral upper-limb fine motor imagination paradigm and the collection of data from 20 healthy people. It presents a feature extraction algorithm for multi-domain fusion and compares the common spatial pattern (CSP), improved multiscale permutation entropy (IMPE) and multi-domain fusion features of all participants through the use of decision tree, linear discriminant analysis, naive Bayes, a support vector machine, k-nearest neighbor and ensemble classification precision algorithms in the ensemble classifier. For the same subject, the average classification accuracy improvement of the same classifier for multi-domain feature extraction relative to CSP feature results went up by 1.52%. The average classification accuracy improvement of the same classifier went up by 32.87% relative to the IMPE feature classification results. This study's unilateral fine motor imagery paradigm and multi-domain feature fusion algorithm provide new ideas for upper limb rehabilitation after stroke.

An optimistic firefly algorithm-based deep learning approach for sentiment analysis of COVID-19 tweets.

The unprecedented rise in the number of COVID-19 cases has drawn global attention, as it has caused an adverse impact on the lives of people all over the world. As of December 31, 2021, more than 2, 86, 901, 222 people have been infected with COVID-19. The rise in the number of COVID-19 cases and deaths across the world has caused fear, anxiety and depression among individuals. Social media is the most dominant tool that disturbed human life during this pandemic. Among the social media platforms, Twitter is one of the most prominent and trusted social media platforms. To control and monitor the COVID-19 infection, it is necessary to analyze the sentiments of people expressed on their social media platforms. In this study, we proposed a deep learning approach known as a long short-term memory (LSTM) model for the analysis of tweets related to COVID-19 as positive or negative sentiments. In addition, the proposed approach makes use of the firefly algorithm to enhance the overall performance of the model. Further, the performance of the proposed model, along with other state-of-the-art ensemble and machine learning models, has been evaluated by using performance metrics such as accuracy, precision, recall, the AUC-ROC and the F1-score. The experimental results reveal that the proposed LSTM + Firefly approach obtained a better accuracy of 99.59% when compared with the other state-of-the-art models.

Identification of immune subtypes of melanoma based on single-cell and bulk RNA sequencing data.

The tumor microenvironment plays a crucial role in melanoma. In this study, the abundance of immune cells in melanoma samples was assessed and analyzed using single sample gene set enrichment analysis (ssGSEA), and the predictive value of immune cells was assessed using univariate COX regression analysis. The Least Absolute Shrinkage and Selection Operator (LASSO)-Cox regression analysis was applied to construct an immune cell risk score (ICRS) model with a high predictive value for identifying the immune profile of melanoma patients. The pathway enrichment between the different ICRS groups was also elucidated. Next, five hub genes for diagnosing the prognosis of melanoma were screened by two machine learning algorithms, LASSO and random forest. The distribution of hub genes in immune cells was analyzed on account of Single-cell RNA sequencing (scRNA-seq), and the interaction between genes and immune cells was elucidated by cellular communication. Ultimately, the ICRS model on account of two types of immune cells (Activated CD8 T cell and Immature B cell) was constructed and validated, which can determine melanoma prognosis. In addition, five hub genes were identified as potential therapeutic targets affecting the prognosis of melanoma patients.

Applying machine learning techniques to detect the deployment of spatial working memory from the spiking activity of MT neurons.

Neural signatures of working memory have been frequently identified in the spiking activity of different brain areas. However, some studies reported no memory-related change in the spiking activity of the middle temporal (MT) area in the visual cortex. However, recently it was shown that the content of working memory is reflected as an increase in the dimensionality of the average spiking activity of the MT neurons. This study aimed to find the features that can reveal memory-related changes with the help of machine-learning algorithms. In this regard, different linear and nonlinear features were obtained from the neuronal spiking activity during the presence and absence of working memory. To select the optimum features, the Genetic algorithm, Particle Swarm Optimization, and Ant Colony Optimization methods were employed. The classification was performed using the Support Vector Machine (SVM) and the K-Nearest Neighbor (KNN) classifiers. Our results suggest that the deployment of spatial working memory can be perfectly detected from spiking patterns of MT neurons with an accuracy of 99.65±0.12 using the KNN and 99.50±0.26 using the SVM classifiers.

Research on cell detection method for microfluidic single cell dispensing.

Single cell dispensing techniques mainly include limiting dilution, fluorescent-activated cell sorting (FACS) and microfluidic approaches. Limiting dilution process is complicated by statistical analysis of clonally derived cell lines. Flow cytometry and conventional microfluidic chip methods utilize excitation fluorescence signals for detection, potentially causing a non-negligible effect on cell activity. In this paper, we implement a nearly non-destructive single-cell dispensing method based on object detection algorithm. To realize single cell detection, we have built automated image acquisition system and then employed PP-YOLO neural network model as detection framework. Through architecture comparison and parameter optimization, we select ResNet-18vd as backbone for feature extraction. We train and evaluate the flow cell detection model on train and test set consisting of 4076 and 453 annotated images respectively. Experiments show that the model inference an image of 320 × 320 pixels at least 0.9 ms with the precision of 98.6% on a NVidia A100 GPU, achieving a good balance of detection speed and accuracy.

Identification of an endoplasmic reticulum stress-related gene signature to predict prognosis and potential drugs of uterine corpus endometrial cancer.

Uterine corpus endometrial cancer (UCEC) is the sixth most common female cancer worldwide, with an increasing incidence. Improving the prognosis of patients living with UCEC is a top priority. Endoplasmic reticulum (ER) stress has been reported to be involved in tumor malignant behaviors and therapy resistance, but its prognostic value in UCEC has been rarely investigated. The present study aimed to construct an ER stress-related gene signature for risk stratification and prognosis prediction in UCEC. The clinical and RNA sequencing data of 523 UCEC patients were extracted from TCGA database and were randomly assigned into a test group (n = 260) and training group (n = 263). An ER stress-related gene signature was established by LASSO and multivariate Cox regression in the training group and validated by Kaplan-Meier survival analysis, Receiver Operating Characteristic (ROC) curves and nomograms in the test group. Tumor immune microenvironment was analyzed by CIBERSORT algorithm and single-sample gene set enrichment analysis. R packages and the Connectivity Map database were used to screen the sensitive drugs. Four ERGs (ATP2C2, CIRBP, CRELD2 and DRD2) were selected to build the risk model. The high-risk group had significantly reduced overall survival (OS) (P < 0.05). The risk model had better prognostic accuracy than clinical factors. Tumor-infiltrating immune cells analysis depicted that CD8+ T cells and regulatory T cells were more abundant in the low-risk group, which may be related to better OS, while activated dendritic cells were active in the high-risk group and associated with unfavorable OS. Several kinds of drugs sensitive to the high-risk group were screened out. The present study constructed an ER stress-related gene signature, which has the potential to predict the prognosis of UCEC patients and have implications for UCEC treatment.

Preimplantation Genetic Testing for Aneuploidy (PGT-A) Reveals High Levels of Chromosomal Errors in In Vivo-Derived Pig Embryos, with an Increased Incidence When Produced In Vitro.

Preimplantation genetic testing for aneuploidy (PGT-A) is widespread, but controversial, in humans and improves pregnancy and live birth rates in cattle. In pigs, it presents a possible solution to improve in vitro embryo production (IVP), however, the incidence and origin of chromosomal errors remains under-explored. To address this, we used single nucleotide polymorphism (SNP)-based PGT-A algorithms in 101 in vivo-derived (IVD) and 64 IVP porcine embryos. More errors were observed in IVP vs. IVD blastocysts (79.7% vs. 13.6% p < 0.001). In IVD embryos, fewer errors were found at blastocyst stage compared to cleavage (4-cell) stage (13.6% vs. 40%, p = 0.056). One androgenetic and two parthenogenetic embryos were also identified. Triploidy was the most common error in IVD embryos (15.8%), but only observed at cleavage, not blastocyst stage, followed by whole chromosome aneuploidy (9.9%). In IVP blastocysts, 32.8% were parthenogenetic, 25.0% (hypo-)triploid, 12.5% aneuploid, and 9.4% haploid. Parthenogenetic blastocysts arose from just three out of ten sows, suggesting a possible donor effect. The high incidence of chromosomal abnormalities in general, but in IVP embryos in particular, suggests an explanation for the low success of porcine IVP. The approaches described provide a means of monitoring technical improvements and suggest future application of PGT-A might improve embryo transfer success.

Analysis of the Low-Carbon Transition Effect and Development Pattern of Green Credit for Prefecture-Level Cities in the Yellow River Basin.

Green credit is a vital instrument for promoting low-carbon transition. However, designing a reasonable development pattern and efficiently allocating limited resources has become a challenge for developing countries. The Yellow River Basin, a critical component of the low-carbon transition in China, is still in the early stages of green credit development. Most cities in this region lack green credit development plans that suit their economic conditions. This study examined the impact of green credit on carbon emission intensity and utilized a k-means clustering algorithm to categorize the green credit development patterns of 98 prefecture-level cities in the Yellow River Basin based on four static indicators and four dynamic indicators. Regression results based on city-level panel data from 2006 to 2020 demonstrated that the development of green credit in the Yellow River Basin can effectively reduce local carbon emission intensity and promote low-carbon transition. We classified the development patterns of green credit in the Yellow River Basin into five types: mechanism construction, product innovation, consumer business expansion, rapid growth, and stable growth. Moreover, we have put forward specific policy suggestions for cities with different development patterns. The design process of this green credit development patterns is characterized by its ability to achieve meaningful outcomes while relying on fewer numbers of indicators. Furthermore, this approach boasts a significant degree of explanatory power, which may assist policy makers in comprehending the underlying mechanisms of regional low-carbon governance. Our findings provide a new perspective for the study of sustainable finance.

Clinical Validation of 5 Direct-to-Consumer Wearable Smart Devices to Detect Atrial Fibrillation: BASEL Wearable Study.

BACKGROUND: Multiple smart devices capable to detect atrial fibrillation (AF) are presently available. Sensitivity and specificity for the detection of AF may differ between available smart devices, and this has not yet been adequately investigated. OBJECTIVES: The aim was to assess the accuracy of 5 smart devices in identifying AF compared with a physician-interpreted 12-lead electrocardiogram as the reference standard in a real-world cohort of patients. METHODS: We consecutively enrolled patients presenting to a cardiology service at a tertiary referral center in a prospective, diagnostic study. RESULTS: We prospectively analyzed 201 patients (31% women, median age 66.7 years). AF was present in 62 (31%) patients. Sensitivity and specificity for the detection of AF were comparable between devices: 85% and 75% for the Apple Watch 6, 85% and 75% for the Samsung Galaxy Watch 3, 58% and 75% for the Withings Scanwatch, 66% and 79% for the Fitbit Sense, and 79% and 69% for the AliveCor KardiaMobile, respectively. The rate of inconclusive tracings (the algorithm was unable to determine the heart rhythm) was 18%, 17%, 24%, 21%, and 26% for the Apple Watch 6, Samsung Galaxy Watch 3, Withings Scan Watch, Fitbit Sense, and AliveCor KardiaMobile (P < 0.01 for pairwise comparison), respectively. By manual review of inconclusive tracings, the rhythm could be determined in 955 (99%) of 969 single-lead electrocardiograms. Regarding patient acceptance, the Apple Watch was ranked first (39% of participants). CONCLUSIONS: In this clinical validation of 5 direct-to-consumer smart devices, we found differences in the amount of inconclusive tracings diminishing sensitivity and specificity of the smart devices. In a clinical setting, manual review of tracings is required in about one-fourth of cases.

Development of an AI-Driven QT Correction Algorithm for Patients in Atrial Fibrillation.

BACKGROUND: Prolongation of the QTc interval is associated with the risk of torsades de pointes. Determination of the QTc interval is therefore of critical importance. There is no reliable method for measuring or correcting the QT interval in atrial fibrillation (AF). OBJECTIVES: The authors sought to evaluate the use of a convolutional neural network (CNN) applied to AF electrocardiograms (ECGs) for accurately estimating the QTc interval and ruling out prolongation of the QTc interval. METHODS: The authors identified patients with a 12-lead ECG in AF within 10 days of a sinus ECG, with similar (±10 ms) QRS durations, between October 23, 2001, and November 5, 2021. A multilayered deep CNN was implemented in TensorFlow 2.5 (Google) to predict the MUSE (GE Healthcare) software-generated sinus QTc value from an AF ECG waveform, demographic characteristics, and software-generated features. RESULTS: The study identified 6,432 patients (44% female) with an average age of 71 years. The CNN predicted sinus QTc values with a mean absolute error of 22.2 ms and root mean squared error of 30.6 ms, similar to the intrinsic variability of the sinus QTc interval. Approximately 84% and 97% of the model's predictions were contained within 1 SD (±30.6 ms) and 2 SD (±61.2 ms) from the sinus QTc interval. The model outperformed the AFQTc method, exhibiting narrower error ranges (mean absolute error comparison P < 0.0001). The model performed best for ruling out QTc prolongation (negative predictive value 0.82 male, 0.92 female; specificity 0.92 male, 0.97 female). CONCLUSIONS: A CNN model applied to AF ECGs accurately predicted the sinus QTc interval, outperforming current alternatives and exhibiting a high negative predictive value.