Artificial Intelligence Deep Learning Research Articles

Using Convolutional Neural Networks in the Development of a Water Pipe Leakage and Location Identification System

To overcome the challenges brought about by abnormal weather and the growing industrial water consumption in Taiwan, the Taiwanese government is transporting water from the northern to the southern part of the country to help with droughts occurring in Taoyuan and Hsinchu. In addition, the government invested NTD 2.78 billion to build the backup water pipelines necessary in Taiyuan and Hsinchu, which help ensure a stable and safe water supply required for regional economic development. The construction adheres to the four major strategic goals of “open source, throttling, dispatch, and backup”. However, the leakage rate of water pipelines remains high. To help with large-scale right-of-way applications and the timeliness of emergency repairs, establishing a system that can detect the locations of leakages is vital. This study intended to apply artificial intelligence (AI) deep learning to develop a water pipe leakage and location identification system. This research established an intelligent sound-assisted water leak identification system, developed and used a localized AI water leak diagnostic instrument to capture on-site dynamic audio, and integrated Internet of Things (IoT) technology to simultaneously identify and locate the leakage. Actual excavation verification results show that the accuracy of the convolutional neural network (CNN) after training is greater than 95%, and the average absolute error calculated between the output data and the input data of the encoder is 0.1021, confirming that the system has high reliability and can reduce the cost of excavation by 26%.

Deep learning model to identify homonymous defects on automated perimetry

BackgroundHomonymous visual field (VF) defects are usually an indicator of serious intracranial pathology but may be subtle and difficult to detect. Artificial intelligence (AI) models could play a key role...

Construction of a new automatic grading system for jaw bone mineral density level based on deep learning using cone beam computed tomography

To develop and verify an automatic classification method using artificial intelligence deep learning to determine the bone mineral density level of the implant site in oral implant surgery from radiographic data obtained from cone beam computed tomography (CBCT) images. Seventy patients with mandibular dentition defects were scanned using CBCT. These Digital Imaging and Communications in Medicine data were cut into 605 training sets, and then the data were processed with data standardization, and the Hounsfiled Unit (HU) value level was determined as follows: Type 1, 1000–2000; type 2, 700–1000; type 3, 400–700; type 4, 100–400; and type 5, − 200–100. Four trained dental implant physicians manually identified and classified the area of the jaw bone density level in the image using the software LabelMe. Then, with the assistance of the HU value generated by LabelMe, a physician with 20 years of clinical experience confirmed the labeling level. Finally, the HU mean values of various categories marked by dental implant physicians were compared to the mean values detected by the artificial intelligence model to assess the accuracy of artificial intelligence classification. After the model was trained on 605 training sets, the statistical results of the HU mean values of various categories in the dataset detected by the model were almost the same as the HU grading interval on the data annotation. This new classification provides a more detailed solution to guide surgeons to adjust the drilling rate and tool selection during preoperative decision-making and intraoperative hole preparation for oral implant surgery.

High-Fidelity Digital Twin Data Models by Randomized Dynamic Mode Decomposition and Deep Learning with Applications in Fluid Dynamics

The purpose of this paper is the identification of high-fidelity digital twin data models from numerical code outputs by non-intrusive techniques (i.e., not requiring Galerkin projection of the governing equations onto the reduced modes basis). In this paper the author defines the concept of the digital twin data model (DTM) as a model of reduced complexity that has the main feature of mirroring the original process behavior. The significant advantage of a DTM is to reproduce the dynamics with high accuracy and reduced costs in CPU time and hardware for settings difficult to explore because of the complexity of the dynamics over time. This paper introduces a new framework for creating efficient digital twin data models by combining two state-of-the-art tools: randomized dynamic mode decomposition and deep learning artificial intelligence. It is shown that the outputs are consistent with the original source data with the advantage of reduced complexity. The DTMs are investigated in the numerical simulation of three shock wave phenomena with increasing complexity. The author performs a thorough assessment of the performance of the new digital twin data models in terms of numerical accuracy and computational efficiency.

Predicting Systemic Health Features from Retinal Fundus Images Using Transfer-Learning-Based Artificial Intelligence Models.

While color fundus photos are used in routine clinical practice to diagnose ophthalmic conditions, evidence suggests that ocular imaging contains valuable information regarding the systemic health features of patients. These features can be identified through computer vision techniques including deep learning (DL) artificial intelligence (AI) models. We aim to construct a DL model that can predict systemic features from fundus images and to determine the optimal method of model construction for this task. Data were collected from a cohort of patients undergoing diabetic retinopathy screening between March 2020 and March 2021. Two models were created for each of 12 systemic health features based on the DenseNet201 architecture: one utilizing transfer learning with images from ImageNet and another from 35,126 fundus images. Here, 1277 fundus images were used to train the AI models. Area under the receiver operating characteristics curve (AUROC) scores were used to compare the model performance. Models utilizing the ImageNet transfer learning data were superior to those using retinal images for transfer learning (mean AUROC 0.78 vs. 0.65, p-value < 0.001). Models using ImageNet pretraining were able to predict systemic features including ethnicity (AUROC 0.93), age > 70 (AUROC 0.90), gender (AUROC 0.85), ACE inhibitor (AUROC 0.82), and ARB medication use (AUROC 0.78). We conclude that fundus images contain valuable information about the systemic characteristics of a patient. To optimize DL model performance, we recommend that even domain specific models consider using transfer learning from more generalized image sets to improve accuracy.

Cultural and Creative Product Design and Image Recognition Based on Deep Learning.

In today's technological world, advanced intelligence technologies such as deep learning (DL) techniques are widely applied in various fields. In this study, people are going to research cultural and creative product design and image recognition based on deep learning. Cultural creative products are referred to as products that are designed by taking inspiration from the cultural aspects. The use of cultural and creative products has increased among the people, thus creating a fair market. Artificial intelligence deep learning (DL) is employed for the design of culturally creative objects. Deep learning is referred to as a machine learning technique that is used to teach machines to imitate human behaviour so that computers can learn from examples. The proposed system utilizes image recognition technique which is referred as the ability of computer systems to identify objects from an image. The image recognition technique integrates machine vision technology, which uses cameras and artificial intelligent software for recognising images. This technology is widely used for various functions, such as self-driven cars, image content searches, and machine vision robots. In our proposed system, image recognition based on deep learning is used in the design of cultural and creative products through the utilisation of randomized algorithms. The system is found to deliver more accurate solutions when compared with the existing LDA, HMM, and optimization algorithms.

Artificial intelligence in ophthalmology: an insight into neurodegenerative disease.

The aging world population accounts for the increasing prevalence of neurodegenerative diseases such as Alzheimer's and Parkinson's which carry a significant health and economic burden. There is therefore a need for sensitive and specific noninvasive biomarkers for early diagnosis and monitoring. Advances in retinal and optic nerve multimodal imaging as well as the development of artificial intelligence deep learning systems (AI-DLS) have heralded a number of promising advances of which ophthalmologists are at the forefront. The association among retinal vascular, nerve fiber layer, and macular findings in neurodegenerative disease is well established. In order to optimize the use of these ophthalmic parameters as biomarkers, validated AI-DLS are required to ensure clinical efficacy and reliability. Varied image acquisition methods and protocols as well as variability in neurogenerative disease diagnosis compromise the robustness of ground truths that are paramount to developing high-quality training datasets. In order to produce effective AI-DLS for the diagnosis and monitoring of neurodegenerative disease, multicenter international collaboration is required to prospectively produce large inclusive datasets, acquired through standardized methods and protocols. With a uniform approach, the efficacy of resultant clinical applications will be maximized.

A Review of Deep Learning-based Human Activity Recognition on Benchmark Video Datasets

ABSTRACT Different types of research have been done on video data using Artificial Intelligence (AI) deep learning techniques. Most of them are behavior analysis, scene understanding, scene labeling, human activity recognition (HAR), object localization, and event recognition. Among all these, HAR is one of the challenging tasks and thrust areas of video data processing research. HAR is applicable in different areas, such as video surveillance systems, human-computer interaction, human behavior characterization, and robotics. This paper aims to present a comparative review of vision-based human activity recognition with the main focus on deep learning techniques on various benchmark video datasets comprehensively. We propose a new taxonomy for categorizing the literature as CNN and RNN-based approaches. We further divide these approaches into four sub-categories and present various methodologies with their experimental datasets and efficiency. A short comparison is also made with the handcrafted feature-based approach and its fusion with deep learning to show the evolution of HAR methods. Finally, we discuss future research directions and some open challenges on human activity recognition. The objective of this survey is to give the current progress of vision-based deep learning HAR methods with the up-to-date study of literature.

High-resolution imaging of subsurface infrastructure using deep learning artificial intelligence on drone magnetometry

The use of drones fo r geophysical data acquisition and artificial intelligence (AI) for geophysical data processing, imaging, and interpretation are active focus areas in current industry and academic applications. Unlocking their cumulative potential in single-focus applications can have a transformative impact, possibly leading to dramatic cost reductions in key use cases and new application areas for enhanced actionable business intelligence. We present field study results from Texas and California that show the potential for imaging pipelines and other subsurface infrastructure by using AI-based methods on high-resolution aboveground magnetic data. The superior resolution and interpretability over conventional geophysical inversion is demonstrated. The method has the potential to provide actionable intelligence in several business-use cases for detecting and characterizing pipelines, crossing zones for multiple pipes, etc. at dramatically reduced costs. The advanced algorithms and workflows used resulted in a 100-fold increase in efficiency and delivered results in two days compared to what could take several months using generally available open-source deep learning AI workflows and software. Future direction of development is to validate against excavation-/drill-bit-/inline-tool-based ground truth and further extend and develop this process to deliver near real-time results. The techniques used are general and can be applied to other geophysical data including seismic, electromagnetic, and gravity at various scales and resolution.

AI-based prediction of new binding site and virtual screening for the discovery of novel P2X3 receptor antagonists

Artificial intelligence (AI) has been recognized as a powerful technique that can accelerate drug discovery during the hit compound identification step. However, most simple deep learning models have been used for naive pre-filtering as the prediction result cannot be interpreted. Recently, our group developed a new deep learning model (Highlight on Target Sequence; HoTS) that can predict binding regions in a target protein sequence based on patterns learned from interactions between a target protein sequence and a ligand. In this study, we searched for new binding regions of the P2X3 receptor (P2X3R) using HoTS, and suggested a novel putative binding site of P2X3R by a cavity search on the predicted binding regions. The novel putative binding site was employed to generate pharmacophore features, and combinations of pharmacophore features were validated as queries. Two separate virtual screenings using the optimized pharmacophore query Q12 with docking-based scoring and HoTS-based prediction of ligand interactions enabled the initial selection of the compound library for in vitro screening. The screening of each set of 500 compounds from the two approaches (HoTS interaction prediction and Pharmacophore-LibDock cascade) resulted in the identification of 10 (HoTS-1 – 10) and 6 compounds (PD-1 – 6) with low micromolar IC50 values. Remarkably, the hit rate was 10-fold higher than that from the previous random screening of 8364 compound library, and the chemical structures of all identified hit compounds were distinct from those of known P2X3R antagonists, indicating that novel chemical entities could be developed for P2X3R antagonists by targeting the binding site. Overall, this study suggests the discovery of a novel putative binding site for P2X3R using the AI deep learning protocol along with in silico MD simulation and experimental screening of targeted library compounds to successfully identify 16 unique and novel hit compounds. These results may accelerate the discovery of novel chemical-class drugs for P2X3R antagonists.

On the Automatic Detection and Classification of Skin Cancer Using Deep Transfer Learning.

Skin cancer (melanoma and non-melanoma) is one of the most common cancer types and leads to hundreds of thousands of yearly deaths worldwide. It manifests itself through abnormal growth of skin cells. Early diagnosis drastically increases the chances of recovery. Moreover, it may render surgical, radiographic, or chemical therapies unnecessary or lessen their overall usage. Thus, healthcare costs can be reduced. The process of diagnosing skin cancer starts with dermoscopy, which inspects the general shape, size, and color characteristics of skin lesions, and suspected lesions undergo further sampling and lab tests for confirmation. Image-based diagnosis has undergone great advances recently due to the rise of deep learning artificial intelligence. The work in this paper examines the applicability of raw deep transfer learning in classifying images of skin lesions into seven possible categories. Using the HAM1000 dataset of dermoscopy images, a system that accepts these images as input without explicit feature extraction or preprocessing was developed using 13 deep transfer learning models. Extensive evaluation revealed the advantages and shortcomings of such a method. Although some cancer types were correctly classified with high accuracy, the imbalance of the dataset, the small number of images in some categories, and the large number of classes reduced the best overall accuracy to 82.9%.

A comparison of artificial intelligence versus radiologists in the diagnosis of thyroid nodules using ultrasonography: a systematic review and meta-analysis.

Thyroid nodules are common. Ultrasonography (US) is the first investigation for thyroid nodules. Artificial Intelligence (AI) is widely integrated into medical diagnosis to provide additional information. The primary objective of this study was to accumulate the pooled sensitivity and specificity between all available AI and radiologists using thyroid US imaging. The secondary objective was to compare AI's diagnostic performance to that of radiologists. A systematic review meta-analysis. PubMed, Scopus, Web of Science, and Cochrane Library data were searched for studies from inception until June 11, 2020. Twenty five studies were included in this meta-analysis. The pooled sensitivity and specificity of AI were 0.86 (95% CI 0.81-0.91) and 0.78 (95% CI 0.73-0.83), respectively. The pooled sensitivity and specificity of radiologists were 0.85 (95% CI 0.80-0.89) and 0.82 (95% CI 0.77-0.86), respectively. The accuracy of AI and radiologists is equivalent in terms of AUC [AI 0.89 (95% CI 0.86-0.92), radiologist 0.91 (95% CI 0.88-0.93)]. The diagnostic odd ratio (DOR) between AI 23.10 (95% CI 14.20-37.58) and radiologists 27.12 (95% CI 17.45-42.16) had no statistically significant difference (P = 0.56). Meta-regression analysis revealed that Deep Learning AI had significantly greater sensitivity and specificity than classic machine learning AI (P < 0.001). AI demonstrated comparable performance to radiologists in diagnosing benign and malignant thyroid nodules using ultrasonography. Additional research to establish its equivalency should be conducted.

P-166 Beyond deep learning: automatic morpho-kinetic parameter extraction using spatial attention

Abstract Study question Can interpretable deep learning Artificial Intelligence automatically learn to extract morphokinetic parameters from time-lapse embryo videos? Summary answer We trained several popular interpretable AI architectures that were successfully able to recognize most embryo developmental stages from day 1 to 5. What is known already Several deep learning AI approaches have been applied to IVF, leading to promising but heterogeneous and perfectible results. However, none of these « black box » models are interpretable, preventing users from having access to the features included in the model and thus from critically analyzing its results. Some interpretable AIs, have been recently developed and could help users visualize the regions of interest in a picture. These interpretable AI have not been adapted and tested to embryo IVF images/videos but could be extremely relevant in this field, providing more insight into AI functioning, and allowing critical analysis. Study design, size, duration In this monocentric study we used 756 randomly selected time lapse videos recorded between 2011 and 2016 (approximately 10% of all the videos generated over this period) and used half of the dataset for the training phase of the AI and the other half for the final evaluation. Participants/materials, setting, methods We used popular deep learning architectures augmented with spatial attention modules that allow the AI to focus on specific areas of the image. We used ResNet-50 as the principal model and combined it with B-CNN, BR-NPA, ProtoPNet, and ProtoTree attention modules. We also applied state-of-the-art generic explanation methods like Grad-CAM, Grad-CAM ++, or Score-CAM. We used standard training procedures and visualized the explanation map produced with a heatmap. Main results and the role of chance We trained several popular interpretable AI architectures to distinguish the 16 development phases of embryos from day 1 to 5 and obtained overall 70 % accuracy, as compared to manual annotation by experts. The inspection of the explanation maps highlighted various patterns of focus on the embryo of one particular model, BR-NPA. At the fertilization stage, this interpretable AI focused on the pro-nuclei, whereas from phase t5 to t9+ the model focused on the separate cells. At the blastocyst stage, the model was able to separate the ICM from the rest of the embryo (trophectoderm). This shows that interpretable AI can indeed be applied to time-lapse videos and focus on relevant regions of the embryo at all developmental stages, producing interpretable results. The role of chance was minimal in this study as we obtained consistent results across various AI architectures, the dataset was selected randomly among all available videos and the dataset was large enough to ensure minimal selection bias. Limitations, reasons for caution Our study suffers from the inherent limitations of monocentric studies. Indeed, lab procedures and characteristics of the population can vary from one setting to another and question the generalizability of our findings. However, deep learning is versatile and similar results can probably be obtained on data from other centers. Wider implications of the findings We postulate that this original approach will help improve the acceptability and the trust of embryologists in AI-based software, which will improve data handling and consistency, ultimately benefiting infertile patients with improved clinical success rates. Trial registration number Not applicable

Privacy Protection Algorithm Model of We-Media Network from the Perspective of Big Data

In recent years, with the in-depth research on the privacy protection methods of We-media network, various types of big data analysis technologies gradually protect the network privacy data, but there are still problems of low intelligence and poor protection in the existing research. Based on this, this paper first uses big data technology and artificial intelligence deep learning algorithm to complete the construction of different types of self-media control databases. Then, it analyzes the common privacy data types of We-media network, constructs an optimization protection model based on secondary identification and verification strategy, and forms a data query system. Finally, simulation experiments are conducted to verify whether the constructed network privacy protection model can realize the intelligent protection of network privacy algorithms from different dimensions. In the process of privacy protection of experimental data at different stages, the internal correlation differences of different types of protection algorithm strategies are obvious in the multidimensional analysis of specific databases. For different types of factor data in different types of We-media networks, the protection rate of the We-media network privacy protection model designed in this study has reached more than 95%. The research results show that the self-media network privacy protection model based on big data and artificial intelligence deep learning technology can realize protection from the aspects of gateway verification and data encryption and has high accuracy and reliability.

Using deep learning artificial intelligence and multiobjective optimization in obtaining the optimum ratio of a fuel cell to electrolyzer power in a hydrogen storage system

Using deep learning artificial intelligence and multiobjective optimization in obtaining the optimum ratio of a fuel cell to electrolyzer power in a hydrogen storage system

Ultra-dense moving cascaded metasurface holography by using a physics-driven neural network.

Metasurfaces are promising platforms for integrated compact optical systems. Traditional metasurface holography design algorithms are limited to information capacity due to finite spatial bandwidth production, which is insufficient for the growing demand for big data storage and encryption. Here, we propose and demonstrate deep learning empowered ultra-dense complex-amplitude holography using step-moving cascaded metasurfaces. Using deep learning artificial intelligence optimization strategy, the barriers of traditional algorithms can be conquered to meet diverse practical requirements. Two metasurfaces are cascaded to form the desired holography. One of them can move to switch the reconstruction images due to diffraction propagation accumulated during the cascaded path. The diffraction pattern from the first metasurface propagates at a different distance and meets with the second metasurface, reconstructing the target holographic reconstructions in the far-field. Such a technique can provide a new solution for multi-dimensional beam shaping, optical encryption, camouflage, integrated on-chip ultra-high-density storage, etc.

Influence Analysis of Education Policy on Migrant Children's Education Integration Using Artificial Intelligence and Deep Learning.

This work intends to solve the problem that the traditional education system cannot reasonably adjust the educational integration of children with the arrival of labor force in a short time, and support the education of migrant children (MC) in the education policy (EP) to integrate them into the local educational environment as soon as possible. Firstly, this work defines the surplus labor force and MC. Secondly, the principles of Artificial Intelligence (AI) and Deep Learning (DL) are introduced. Thirdly, it analyzes the education of MC and relevant policies, and the data of the education effect of MC are collected and the evaluation effect model is built. Finally, the evaluation model of MC’s education effect is applied to test the effect of EP. The results show that using AI technology combined with DL technology to model the education effect of MC can establish an effective and accurate evaluation model of the education effect of MC, effectively evaluate the impact of local education policies on the education of MC, and give an effective effect analysis of relevant education policies in each period. The result of Adaptive Resonance Theory (ART)–Back Propagation algorithm is 65 ∼ 96%, which is much higher than the efficiency of traditional algorithms. This shows that the education integration evaluation model of MC based on AI technology and DL technology can effectively and accurately evaluate the integration effect of MC on the local education system, and then provide reference for local and even national adjustment of education policies. The results provide a new idea for the application of new technology in EP.

Effect of 320-row CT reconstruction technology on fractional flow reserve derived from coronary CT angiography based on machine learning: single- versus multiple-cardiac periodic images.

Fractional flow reserve derived from computed tomography (CT-FFR) can be used to noninvasively evaluate the functions of coronary arteries and has been widely welcomed in the field of cardiovascular research. However, whether different image reconstruction schemes have an effect on CT-FFR analysis through single- and multiple-cardiac periodic images in the same patient has not been investigated. This study retrospectively enrolled 122 patients who underwent 320-row computed tomography (CT) examination with both single- and multiple-cardiac periodic reconstruction schemes; a total of 366 coronary arteries were analyzed. The lowest CT-FFR values of each vessel and the poststenosis CT-FFR values of the lesion-specific coronary artery were measured using the two reconstruction techniques. The Wilcoxon signed-rank test was used to compare differences in CT-FFR values between the two reconstruction techniques. Spearman correlation analysis was performed to determine the relationship between CT-FFR values derived using the two methods. Bland-Altman and intraclass correlation coefficient (ICC) analyses were performed to evaluate the consistency of CT-FFR values. In all blood vessels, the lowest CT-FFR values showed no significant differences between the two reconstruction techniques in the left anterior descending artery (LAD; P=0.65), left circumflex artery (LCx; P=0.46), or right coronary artery (RCA; P=0.22). In blood vessels with atherosclerotic plaques, the poststenosis CT-FFR values (2 cm distal to the maximum stenosis) exhibited no significant differences between the two reconstruction techniques in the LAD (P=0.78), LCx (P=1.00), or RCA (P=1.00). The mean CT-FFR values of single- and multiple-cardiac periodic images showed excellent correlation and minimal bias in all groups. CT-FFR analysis based on an artificial intelligence deep learning neural network is stable and not affected by the type of 320-row CT reconstruction technology.

Patient and procedure selection for mechanical thrombectomy: Toward personalized medicine and the role of artificial intelligence.

Mechanical thrombectomy (MT) for ischemic stroke due to large vessel occlusion is standard of care. Evidence-based guidelines on eligibility for MT have been outlined and evidence to extend the treatment benefit to more patients, particularly those at the extreme ends of a stroke clinical severity spectrum, is currently awaited. As patient selection continues to be explored, there is growing focus on procedure selection including the tools and techniques of thrombectomy and associated outcomes. Artificial intelligence (AI) has been instrumental in the area of patient selection for MT with a role in diagnosis and delivery of acute stroke care. Machine learning algorithms have been developed to detect cerebral ischemia and early infarct core, presence of large vessel occlusion, and perfusion deficit in acute ischemic stroke. Several available deep learning AI applications provide ready visualization and interpretation of cervical and cerebral arteries. Further enhancement of AI techniques to potentially include automated vessel probe tools in suspected large vessel occlusions is proposed. Value of AI may be extended to assist in procedure selection including both the tools and technique of thrombectomy. Delivering personalized medicine is the wave of the future and tailoring the MT treatment to a stroke patient is in line with this trend.

Mango wine making process optimization based on artificial intelligence deep learning technology

AbstractThis paper combines artificial intelligence deep learning technology to optimize the wine making process of mango wine. Moreover, in view of the shortcomings of traditional electronic nose data processing methods, a deep learning method based on SSAE‐BPNN is proposed for electronic nose data processing. In addition, according to the characteristics of automatic learning features, this paper uses a deep learning method based on SSAE‐BPNN to simplify the process of traditional data processing methods. Finally, this paper constructs an electronic nose system that can be used to identify mango wine making characteristics, and enhances the effect of electronic nose recognition through deep learning. Through the analysis, it can be seen that the mango wine making process optimization method based on artificial intelligence deep learning technology proposed in this paper has a certain effect, and it has optimized the traditional mango wine making process.