Deep Learning Algorithms Research Articles

Plaque phenotyping using deep learning in optical coherence tomography imaging

Abstract Background Artificial intelligence (AI) techniques have emerged as powerful tools in various computer vision tasks, particularly in classification, detection, and segmentation. Purpose This study aimed to develop a deep learning model for categorizing optical coherence tomography (OCT) frames into different plaque phenotypes, including thin-cap fibroatheroma (TCFA) and fibroatheroma. Methods A prospective multicenter observational study enrolled 168 patients who underwent percutaneous coronary intervention (PCI) and received OCT imaging for non-target lesions with angiographically estimated diameter stenosis greater than 30% (177 vessels). Among them, 129 patients (135 vessels) underwent follow-up OCT imaging at 1 year. A total of 45,072 OCT frames were randomly divided into training and validation sets in a 4:1 ratio. An EfficientNet-B3 convolutional neural network model was developed to classify OCT frames into TCFA, thick-cap fibroatheroma, non-fibroatheroma, adaptive intimal thickening, and normal vessel categories. Results A confusion matrix comparing manually assessed plaque phenotypes with AI-predicted phenotypes was presented in Table. In the validation datasets (9,015 frames), the overall accuracy of the model was 85.1%. For predicting the presence of TCFA, sensitivity was 81.1%, specificity was 98.8%, positive predictive value was 63.9%, and negative predictive value was 99.5%. Regarding the presence of fibroatheroma, sensitivity was 90.3%, specificity was 89.4%, positive predictive value was 93.1%, and negative predictive value was 85.4%. Conclusions The deep learning algorithm demonstrated high accuracy in detecting TCFA and fibroatheroma with excellent reproducibility. This automated process enables operators to promptly identify high-risk lesions during PCI, potentially leading to improved clinical outcomes.

A 2D echocardiographic method for accurate right ventricular quantification using deep learning

Abstract Background/Introduction Transthoracic two-dimensional echocardiography (2DE) remains the backbone for non-invasive assessment of the right ventricle (RV) size and function. However, calculation of RV end-diastolic (RVEDV), end-systolic volume (RVESV) and ejection fraction (RVEF) cannot be performed using 2DE due to the complex RV shape (1), whereas currently used, respective 2DE indices lack diagnostic accuracy. Deep learning (DL) may be able to assist in developing new 2DE methods for RV quantification. Purpose To develop a novel DL-based 2DE method for quantitative evaluation of RV size and function without geometric assumptions. Methods 50 adult patients underwent 2DE and CMR within 30 days as part of routine cardiac care between 6/2020 – 9/2021 at a medical center. We excluded patients with prior cardiac surgery of the RV, tricuspid or pulmonic valve repair or replacement, complex congenital heart disease, more than small pericardial effusion, cardiac tamponade, atrial fibrillation at the time of either study, severe RV dysfunction or inadequate imaging quality. End-diastolic and end-systolic areas derived from planimetry of 8 2DE RV views, along with patient age and gender, were to used to train and test a DL algorithm, namely the Feature-Tokenizer and Transformer (Figure 1) in order to predict the corresponding, CMR-derived, RVEDV or RVESV, and subsequently calculate RVEF. We used 5-fold cross-validation for training and evaluation. We assessed the relative importance of each 2DE view using gain metric-powered explainability analysis and subsequently performed a sensitivity analysis to identify optimal combinations of views. Variability analysis was performed for a random subcohort of 10 patients. Results Median age was 51 (interquartile range 32-62) and 42% were women. Mean RVEDV, RVESV and RVEF) by CMR were 163±70ml, 82±42ml and 51±8% respectively. Six patients (12%) had RV dilatation and 9 (18%) had RV dysfunction when using published CMR criteria. Explainability analysis showed that the 3 most important views were parasternal long axis (PLAX) (relative feature importance [RFI] = 0.197), 4-chamber (RFI=0.145) and parasternal short axis at the base (PSAX-base) (RFI=0.112). The 3-view combination of PLAX, 4-chamber and PSAX-base achieved high accuracy when compared to CMR (R2=0.964, absolute percentage error [APE]=8.09±12.07% for RV volumes and APE=9.86±10.68% for RVEF). Good (intraclass correlation coefficient [ICC]>0.75) to excellent (ICC>0.90) intra- and interobserver variability was found for these 3 views. Conclusions We propose an innovative, clinically applicable DL-based 2DE method for quantification of RV size and function that requires manual RV planimetry of 3 views, namely 4-chamber, PLAX and PSAX-base (Figure 2). This method may revolutionize 2DE RV assessment. Further studies with largest number of patients and wider range of RV size and function are needed to validate this method.Deep learning architectureGraphical abstract

Current application of artificial intelligence in laparoscopic cholecystectomy

Recent advances in artificial intelligence (AI) have sparked a surge in the application of computer vision (CV) in surgical video analysis. Surgical complications often occur due to lapses in judgment and decision-making. In laparoscopic cholecystectomy, achievement of the critical view of safety is commonly advocated to prevent bile duct injuries. However, bile duct injuries rates remain stable, probably due to inconsistent application or a poor understanding of critical view of safety. Advances in AI have made it possible to train algorithms that identify anatomy and interpret the surgical field. AI-based CV techniques may leverage surgical video data to develop real-time automated decision support tools and surgeon training systems. The effectiveness of CV application in surgical procedures is still under early evaluation. The review considers the commonly used deep learning algorithms in CV and describes their usage in detail in four application scenes, including phase recognition, anatomy detection, instrument detection and action recognition in laparoscopic cholecystectomy. The MedLine, Scopus, and IEEE Xplore databases were searched for publications up to 2024. The keywords used in the search were “laparoscopic cholecystectomy”, “artificial intelligence”. The currently described applications of CV in laparoscopic cholecystectomy are limited. Most current research focus on the identification of workflow and anatomical structure, while the identification of instruments and surgical actions is still awaiting further breakthroughs. Future research on the use of CV in laparoscopic cholecystectomy should focus on application in more scenarios, such as surgeon skill assessment and the development of more efficient models.

Research and Experiment on Miss-Seeding Detection of Potato Planter Based on Improved YOLOv5s

In order to improve the performance of potato planter, reduce miss-seeding rates, enhance the overall quality of the seeding operation, and ultimately increase the yield of the potato, it is necessary to implement effective technical means to monitor and identify the miss-seeding issues during the seeding process. The existing miss-seeding detection technologies commonly use sensors to monitor, but such technologies are easily affected by factors like heavy dust and strong vibrations, resulting in poor interference resistance and adaptability. Therefore, this study aims to explore and apply deep learning algorithms to achieve real-time monitoring of the miss-seeding phenomenon in potato planter during the planting process. Considering both the lightweight of the miss-seeding detection model and its practical deployment, this study selects and adapts the YOLOv5s algorithm to achieve this goal. Firstly, the attention mechanism is integrated into the backbone network to suppress background interference and improve detection accuracy. Secondly, the non-maximum suppression algorithm is improved by replacing the original IoU-NMS with the Soft-NMS algorithm to enhance the bounding box regression rate and reduce missed detections of potato seeds due to background overlap or occlusion. Experimental results show that the accuracy of the improved algorithm in detecting miss-seeding increased from 96.02% to 98.30%, the recall rate increased from 96.31% to 99.40%, and the mean average precision (mAP) improved from 99.12% to 99.40%. The improved model reduces missed and false detections, provides more precise target localization, and is suitable for miss-seeding detection in natural environments for potato planter, providing technical and theoretical support for subsequent intelligent reseeding in potato planter.

Artificial Intelligence-Based Sentinel Lymph Node Metastasis Detection in Cervical Cancer.

Background/objectives: Pathological ultrastaging, an essential part of sentinel lymph node (SLN) mapping, involves serial sectioning and immunohistochemical (IHC) staining in order to reliably detect clinically relevant metastases. However, ultrastaging is labor-intensive, time-consuming, and costly. Deep learning algorithms offer a potential solution by assisting pathologists in efficiently assessing serial sections for metastases, reducing workload and costs while enhancing accuracy. This proof-of-principle study evaluated the effectiveness of a deep learning algorithm for SLN metastasis detection in early-stage cervical cancer. Methods: We retrospectively analyzed whole slide images (WSIs) of hematoxylin and eosin (H&E)-stained SLNs from early-stage cervical cancer patients diagnosed with an SLN metastasis with either H&E or IHC. A CE-IVD certified commercially available deep learning algorithm, initially developed for detection of breast and colon cancer lymph node metastases, was employed off-label to assess its sensitivity in cervical cancer. Results: This study included 21 patients with early-stage cervical cancer, comprising 15 with squamous cell carcinoma, five with adenocarcinoma, and one with clear cell carcinoma. Among these patients, 10 had macrometastases and 11 had micrometastases in at least one SLN. The algorithm was applied to evaluate H&E WSIs of 47 SLN specimens, including 22 that were negative for metastasis, 13 with macrometastases, and 12 with micrometastases in the H&E slides. The algorithm detected all H&E macro- and micrometastases with 100% sensitivity. Conclusions: This proof-of-principle study demonstrated high sensitivity of a deep learning algorithm for detection of clinically relevant SLN metastasis in early-stage cervical cancer, despite being originally developed for adenocarcinomas of the breast and colon. Our findings highlight the potential of leveraging an existing algorithm for use in cervical cancer, warranting further prospective validation in a larger population.

Grammatical Error Correction Detection of English Conversational Pronunciation Under a Deep Learning Algorithm

Grammar correction in spoken English can enhance proficiency. This paper briefly introduces the gate recurrent unit (GRU) algorithm and its appli- cation in English speech recognition and grammatical error correction of speech recognition results. The GRU algorithm was firstly used to recognize English speech, then transform it into a text, and finally correct the English grammar of the text. Additionally, the attention mechanism was incorporated to enhance the performance of grammatical error correction. Subsequently, simulation experiments were conducted. Firstly, speech recognition and grammatical error correction were independently verified. The performance of the proposed algorithm in correcting grammatical errors in spoken English was evaluated using a self-built speech database. The results demonstrated that the proposed GRU-based algorithm yielded the best performance in independent speech recognition, independent grammatical error correction, and the overall spoken grammatical error correction. The contribution of this study lies in using the GRU algorithm to convert speech into text and perform grammar correction on the text, providing an effective reference for grammar correction in English communication.

Comparison of transformer, LSTM and coupled algorithms for soil moisture prediction in shallow-groundwater-level areas with interpretability analysis

Accurate quantification of soil moisture is essential for understanding water and energy exchanges between the atmosphere and the Earth’s surface, as well as for agricultural applications. Predicting soil moisture content is vital for efficient water management, irrigation scheduling, and drought monitoring. Traditional forecasting methods, such as numerical regression models, often struggle due to various influencing factors and poor observation data quality. In contrast, deep learning algorithms, particularly recurrent and convolutional neural networks, show promise in predicting nonlinear data like soil moisture. This study focuses on shallow groundwater regions, using groundwater levels and meteorological data as features while coupling the Transformer model with other neural network structures. We investigate the potential of attention-based neural networks for soil moisture time series prediction. Our findings demonstrate that the Transformer model achieves an average R2 of 0.523 across different time lags, outperforming the LSTM model with an R2 of 0.485. The introduction of LSTM enhances the Transformer’s stability in handling temporal changes. Additionally, we verified the importance of groundwater for soil moisture prediction. This study introduces new methods for soil moisture prediction and offers new insights and recommendations for the development of artificial intelligence technology for soil moisture prediction.

Data-Driven Approaches for State-of-Charge Estimation in Battery Electric Vehicles Using Machine and Deep Learning Techniques

One of the most important functions of the battery management system (BMS) in battery electric vehicle (BEV) applications is to estimate the state of charge (SOC). In this study, several machine and deep learning techniques, such as linear regression, support vector regressors (SVRs), k-nearest neighbor, random forest, extra trees regressor, extreme gradient boosting, random forest combined with gradient boosting, artificial neural networks (ANNs), convolutional neural networks, and long short-term memory (LSTM) networks, are investigated to develop a modeling framework for SOC estimation. The purpose of this study is to improve overall battery performance by examining how BEV operation affects battery deterioration. By using dynamic response simulation of lithium battery electric vehicles (BEVs) and lithium battery packs (LIBs), the proposed research provides realistic training data, enabling more accurate prediction of SOC using data-driven methods, which will have a crucial and effective impact on the safe operation of electric vehicles. The paper evaluates the performance of machine and deep learning algorithms using various metrics, including the R2 Score, median absolute error, mean square error, mean absolute error, and max error. All the simulation tests were performed using MATLAB 2023, Anaconda platform, and COMSOL Multiphysics.

Evaluating the Accuracy of Machine Learning, Deep Learning and Hybrid Algorithms for Flood Routing Calculations

Evaluating the Accuracy of Machine Learning, Deep Learning and Hybrid Algorithms for Flood Routing Calculations

Predictive Diagnostic Model for Early Osteoporosis Detection Using Deep Learning and Multimodal Imaging Data: A Systematic Review and Meta-Analysis

Osteoporosis is a common condition that weakens bones, making them more prone to fractures. Early detection is crucial for preventing fractures and improving patients’ quality of life. However, traditional methods, such as Dual-Energy X-ray Absorptiometry (DXA), often struggle to accurately predict fracture risk and may overlook minor changes in bone structure. This study focuses on developing a predictive model for early osteoporosis detection using deep learning algorithms combined with various imaging techniques, including MRI, CT, and High-Resolution Peripheral Quantitative Computed Tomography (HR-pQCT). A systematic review and meta-analysis of studies published between 2014 and 2024 were conducted, examining the use of deep learning models applied to multimodal imaging data. The meta-analysis highlighted differences in the accuracy and effectiveness of various models, and their performance was measured in terms of accuracy, sensitivity, and specificity, following PRISMA guidelines. The results showed that deep learning models outperformed traditional methods in early osteoporosis detection. The use of multiple imaging techniques provided a more detailed assessment of bone health, allowing the models to identify complex patterns that are difficult for human interpretation. These models demonstrated high accuracy and significant potential for improving clinical decision-making. By integrating deep learning with multimodal imaging, this approach offers a promising solution for enhancing the early detection of osteoporosis. The models tested in this study proved to be highly effective, yielding more accurate fracture risk predictions and enabling earlier interventions. This could lead to better patient outcomes and reduced healthcare costs.

A review on depth perception techniques in organoid images

Organoids are an in vitro model that can simulate the complex structure and function of tissues in vivo. Functions such as classification, screening and trajectory recognition have been realized through organoid image analysis, but there are still problems such as low accuracy in recognition classification and cell tracking. Deep learning algorithm and organoid image fusion analysis are the most advanced organoid image analysis methods. In this paper, the organoid image depth perception technology is investigated and sorted out, the organoid culture mechanism and its application concept in depth perception are introduced, and the key progress of four depth perception algorithms such as organoid image and classification recognition, pattern detection, image segmentation and dynamic tracking are reviewed respectively, and the performance advantages of different depth models are compared and analyzed. In addition, this paper also summarizes the depth perception technology of various organ images from the aspects of depth perception feature learning, model generalization and multiple evaluation parameters, and prospects the development trend of organoids based on deep learning methods in the future, so as to promote the application of depth perception technology in organoid images. It provides an important reference for the academic research and practical application in this field.

Developing Site-Specific Prescription Maps for Sugarcane Weed Control Using High-Spatial-Resolution Images and Light Detection and Ranging (LiDAR)

Sugarcane is a perennial grass species mainly for sugar production and one of the significant crops in Costa Rica, where ideal growing conditions support its cultivation. Weed control is a critical aspect of sugarcane farming, traditionally managed through preventive or corrective mechanical and chemical methods. However, these methods can be time-consuming and costly. This study aimed to develop site-specific, variable rate prescription maps for weed control using remote sensing. High-spatial-resolution images (5 cm) and Light Detection And Ranging (LiDAR) were acquired using a Micasense Rededge-P camera and a DJI L1 sensor mounted on a drone. Precise locations of weeds were collected for calibration and validation. Normalized Difference Vegetation Index derived from multispectral images separated vegetation coverage and soil. A deep learning (DL) algorithm further classified vegetation coverage into sugarcane and weeds. The DL model performed well without overfitting. The classification accuracy was 87% compared to validation samples. The density and average heights of weed patches were extracted from the canopy height model (LiDAR). They were used to derive site-specific prescription maps for weed control. This efficient and precise alternative to traditional methods could optimize weed control, reduce herbicide usage and provide more profitable yield.

Przykłady wdrożeń i przyszłość AI w diagnostyce medycznej

AI is revolutionizing medical diagnostics around the world, innovating in a variety of contexts, from leading US hospitals to facilities in developing countries. Below we present examples of AI implementations in medical diagnostics from different regions, taking into account the effectiveness and results of these solutions and forecasts for the development of this technology. Regarding the future of artificial intelligence in medical diagnostics, the article considered potential innovations such as the development of deep learning algorithms and integration with 5G technologies and the Internet. Attention is paid to the possibilities of further personalization of healthcare and to the challenges related to the need to adapt legal regulations and data management. It also indicates the directions of future research that may contribute to the further development of AI in medical diagnostics and the improvement of the quality of healthcare not only in Poland, but around the world.

An age-group ranking model for facial age estimation

Age prediction has become an important Computer Vision task. Although this task requires the age of an individual to be predicted from a given face, research has shown that it is more intuitive and easier for humans to decide which of two individuals is older than to decide how old an individual is. This work follows this intuition to aid the age prediction of a face by exploiting the age information available from other faces. It goes further to explore the statistical relationships between facial features within age groups to compute age-group ranks for a given face. The resulting age-group rank is low-dimensional and age-discriminatory, thus improving age prediction accuracy when fed into an age predictor. Experiments on publicly available facial ageing datasets (FGnet, PAL, and Adience) reveal the effectiveness of the proposed age-group ranking model when used with traditional Machine learning algorithms as well as Deep Learning algorithms. Cross-dataset validation, a method of training and testing on entirely different datasets, was also employed to further investigate the effectiveness of this method.

Advancing Electric Engineering Education through Immersive Virtual Reality: Deep Learning and Evolutionary Algorithms for Image Stitching and Rectification in Virtual Lab Environments

Virtual Reality (VR) technology has emerged as a transformative tool in education, offering immersive and interactive experiences that enhance learning outcomes. This paper delves into the application of image stitching and rectification techniques to create a VR lab environment, specifically tailored for electrical engineering education. The importance of VR technology in education is explored, highlighting its role in promoting active learning and providing experiential learning opportunities. The primary emphasis of this Paper lies in the smooth incorporation of image stitching algorithms for the creation of panoramic perspectives, along with the implementation of rectification techniques to correct irregular borders within the stitched images. By utilizing Convolutional Neural Networks (CNNs) and Genetic Algorithms (GAs), the proposed approach optimizes the rectification process, resulting in visually cohesive representations. Demonstrating the utilization of the VR lab across a range of situations, such as examining power transfer and creating control panels for water pumps in irrigation initiatives, the immersive setting enables students to delve into intricate systems. The performance of the proposed method was evaluated using various metrics, including mean squared error, peak signal to noise ratio (PSNR), structural similarity index (SSIM), and Fréchet inception distance (FID). the combination of deep learning algorithm specifically (CNN) and optimization algorithm specifically (Genetic algorithm (GA)) led to an increase in the accuracy of the rectified images where the average PSNR reached 23.98, SSIM was 0.8066, and FID was 18.72. Regarding the users’ opinion about the generated environment by stitching and rectifying images, participants demonstrated consistent positive sentiments, with mean scores ranging from 3.65 to 4.03, all above the scale midpoint, and moderate variability indicated by standard deviation values ranging from 1.070 to 1.251, suggesting general favorability with some variation in responses. This experience empowers the users to gain insights and cultivate essential problemsolving abilities at a heightened level. Collaborative learning is facilitated, enabling students to engage in collaborative projects regardless of their physical location. Through the synthesis of image processing techniques and VR technology, this research contributes to the enrichment of educational experiences and the advancement of electrical engineering education.

Latency‐Sensitive Service Function Chains Intelligent Migration in Satellite Communication Driven by Deep Reinforcement Learning

ABSTRACTSatellite communication technology solves the problem that the traditional wired network infrastructure is difficult to achieve global communication coverage. However, factors such as satellite orbits introduce frequent changes to the network topology, and challenges like satellite failures and communication link interruptions are prevalent. In the face of these issues, service function chain (SFC) migration becomes a crucial method for swiftly adjusting SFCs during faults, maintaining service continuity and availability. This article proposes a latency‐sensitive SFC migration algorithm tailored to satellite networks. The algorithm first models the satellite network as a multi‐domain virtual network, capturing the constraints faced during SFC migration. Subsequently, a deep reinforcement learning algorithm integrated attention mechanism is designed to more accurately capture and understand the complex network environment and dynamic satellite network topology and derive optimal SFC migration strategies for superior performance. Finally, through experimentation and evaluation of the deep reinforcement learning‐driven latency‐sensitive service function chain intelligent migration algorithm (LS‐SFCM) in satellite communication, this study validates the effectiveness and superior performance of the algorithm in latency‐sensitive scenarios. It provides a new avenue for enhancing the service quality and efficiency of satellite communication networks.

Diagnostic accuracy of deep learning-based algorithms in laryngoscopy: a systematic review and meta-analysis.

Laryngoscopy is routinely used for suspicious vocal cord lesions with limited performance. Accumulated studies have demonstrated the bright prospect of deep learning in processing medical imaging. In this study, we perform a systematic review and meta-analysis to investigate diagnostic utility of deep learning in laryngoscopy. The study was performed according to the Primary Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. We comprehensively retrieved articles from the PubMed, Scopus, Embase, and Web of Science up to July 14, 2024. Eligible studies with application of deep learning algorithm in laryngoscopy were assessed and enrolled by two independent investigators. The pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio with 95% confidence intervals (CIs) were calculated using a random effects model. We retained 9 eligible studies adding up to 106,175 endoscopic images for the meta-analysis. The pooled sensitivity and specificity to diagnose laryngeal cancer were 0.95(95% CI: 0.85-0.98) and 0.96 (95% CI: 0.91-0.98). The area under the curve of deep learning was 0.99 (95%CI: 0.97-0.99). Deep learning demonstrated excellent diagnostic efficacy in assessing laryngeal cancer with laryngoscope images in current studies, which manifests its potential of aiding endoscopist for laryngeal cancer diagnosis and clinical decision making.

Application of Hybrid Attention Mechanisms in Lithological Classification with Multisource Data: A Case Study from the Altay Orogenic Belt

Multisource data fusion technology integrates the strengths of various data sources, addressing the limitations of relying on a single source. Therefore, it has been widely applied in fields such as lithological classification and mineral exploration. However, traditional deep learning algorithms fail to distinguish the importance of different features effectively during fusion, leading to insufficient focus in the model. To address this issue, this paper introduces a ResHA network based on a hybrid attention mechanism to fuse features from ASTER remote sensing images, geochemical data, and DEM data. A case study was conducted in the Altay Orogenic Belt to demonstrate the lithological classification process. This study explored the impact of the submodule order on the hybrid attention mechanism and compared the results with those of MLP, KNN, RF, and SVM models. The experimental results show that (1) the ResHA network with hybrid attention mechanisms assigned reasonable weights to the feature sets, allowing the model to focus on key features closely related to the task. This resulted in a 7.99% improvement in classification accuracy compared with that of traditional models, significantly increasing the precision of lithological classification. (2) The combination of channel attention followed by spatial attention achieved the highest overall accuracy, 98.06%.

Identification of nitric oxide-mediated necroptosis as the predominant death route in Parkinson’s disease

Parkinson's disease (PD) involves multiple forms of neuronal cell death, but the dominant pathway involved in disease progression remains unclear. This study employed RNA sequencing (RNA-seq) of brain tissue to explore gene expression patterns across different stages of PD. Using the Scaden deep learning algorithm, we predicted neurocyte subtypes and modelled dynamic interactions for five classic cell death pathways to identify the predominant routes of neuronal death during PD progression. Our cell type-specific analysis revealed an increasing shift towards necroptosis, which was strongly correlated with nitric oxide synthase (NOS) expression across most neuronal subtypes. In vitro experiments confirmed that nitric oxide (NO) is a key mediator of necroptosis, leading to nuclear shrinkage and decreased mitochondrial membrane potential via phosphorylation of the PIP1/PIP3/MLKL signalling cascade. Importantly, specific necroptosis inhibitors significantly mitigated neuronal damage in both in vitro and in vivo PD models. Further analysis revealed that NO-mediated necroptosis is prevalent in early-onset Alzheimer's disease (AD) and amyotrophic lateral sclerosis (ALS) and across multiple brain regions but not in brain tumours. Our findings suggest that NO-mediated necroptosis is a critical pathway in PD and other neurodegenerative disorders, providing potential targets for therapeutic intervention.

Deep learning improves physician accuracy in the comprehensive detection of abnormalities on chest X-rays.

Chest X-rays are the most commonly performed medical imaging exam, yet they are often misinterpreted by physicians. Here, we present an FDA-cleared, artificial intelligence (AI) system which uses a deep learning algorithm to assist physicians in the comprehensive detection and localization of abnormalities on chest X-rays. We trained and tested the AI system on a large dataset, assessed generalizability on publicly available data, and evaluated radiologist and non-radiologist physician accuracy when unaided and aided by the AI system. The AI system accurately detected chest X-ray abnormalities (AUC: 0.976, 95% bootstrap CI: 0.975, 0.976) and generalized to a publicly available dataset (AUC: 0.975, 95% bootstrap CI: 0.971, 0.978). Physicians showed significant improvements in detecting abnormalities on chest X-rays when aided by the AI system compared to when unaided (difference in AUC: 0.101, p < 0.001). Non-radiologist physicians detected abnormalities on chest X-ray exams as accurately as radiologists when aided by the AI system and were faster at evaluating chest X-rays when aided compared to unaided. Together, these results show that the AI system is accurate and reduces physician errors in chest X-ray evaluation, which highlights the potential of AI systems to improve access to fast, high-quality radiograph interpretation.