Neural Net Algorithm Research Articles

Deep learning-based automatic segmentation of COVID-19 in chest X-ray images using ensemble neural net sentinel algorithm

IntroductionCOVID-19 has affected billions of people worldwide and has serious public health effects. Automatic image segmentation is crucial to improve clinical decision-making for full assessments of illness management as well as monitoring. It plays a crucial part in detecting, diagnosing, and tracking diseases by allowing for the exact or accurate segmentation of diseased regions. Automatic segmentation and classifying medical images remains the most difficult obstacle in medical diagnosis, however, because of the lack of annotated medical images. MethodsIn this article, we propose the Ensemble Neural Net Sentinel Algorithm (ENNSA), and unique deep learning (DL) method for the segmentation and classification of COVID-19. Chest X-ray dataset is first collected from Kaggle and preprocessed. The features in the processed image are retrieved using the Insistent Grey Level Co-occurrence Matrix (IGLCM). ResultsThe suggested ENNSA outperforms current techniques in terms of accuracy (99.25), precision (100), recall (43), and F1 score (61) regardless of unbalanced datasets, indicating its promise as a diagnostic tool for respiratory diseases. ConclusionIn addition, the segmentation approach is used to separate the chosen features for COVID-19 classification from the chest X-ray images. The investigational outcomes demonstrate that our proposed strategy outperforms currently used methods in classifying COVID-19.

Prediction of stability constants of metal-ligand complexes by machine learning for the design of ligands with optimal metal ion selectivity.

The new LOGKPREDICT program integrates HostDesigner molecular design software with the machine learning (ML) program Chemprop. By supplying HostDesigner with predicted logK values, LOGKPREDICT enhances the computer-aided molecular design process by ranking ligands directly by metal-ligand binding strength. Harnessing reliable experimental data from a historic National Institute of Standards and Technology (NIST) database and data from the International Union of Pure and Applied Chemistry (IUPAC), we train message passing neural net algorithms. The multi-metal NIST-based ML model has a root mean square error (RMSE) of 0.629 ± 0.044 (R2 of 0.960 ± 0.006), while two versions of lanthanide-only IUPAC-based ML models have, respectively, RMSE of 0.764 ± 0.073 (R2 of 0.976 ± 0.005) and 0.757 ± 0.071 (R2 of 0.959 ± 0.007). For relative logK predictions on an out-of-sample set of six ligands, demonstrating metal ion selectivity, the RMSE value reaches a commendably low 0.25. We showcase the use of LOGKPREDICT in identifying ligands with high selectivity for lanthanides in aqueous solutions, a finding supported by recent experimental evidence. We also predict new ligands yet to be verified experimentally. Therefore, our ML models implemented through LOGKPREDICT and interfaced with the ligand design software HostDesigner pave the way for designing new ligands with predetermined selectivity for competing metal ions in an aqueous solution.

The Prediction of Clinical Mastitis in Dairy Cows Based on Milk Yield, Rumination Time, and Milk Electrical Conductivity Using Machine Learning Algorithms

In commercial dairy farms, mastitis is associated with increased antimicrobial use and associated resistance, which may affect milk production. This study aimed to develop sensor-based prediction models for naturally occurring clinical bovine mastitis using nine machine learning algorithms with data from 447 mastitic and 2146 healthy cows obtained from five commercial farms in Northeast China. The variables were related to daily activity, rumination time, and daily milk yield of cows, as well as milk electrical conductivity. Both Z-standardized and non-standardized datasets pertaining to four specific stages of lactation were used to train and test prediction models. For all four subgroups, the Z-standardized dataset yielded better results than those of the non-standardized one, with the multilayer artificial neural net algorithm showing the best performance. Variables of importance had a similar rank in this algorithm, indicating the consistency of these variables as predictors for bovine mastitis in commercial farms with similar automatic systems. Moreover, the peak milk yield (PMY) of mastitic cows was significantly higher than that of healthy cows (p < 0.005), indicating that high-yielding cattle are more prone to mastitis. Our results show that machine learning algorithms are effective tools for predicting mastitis in dairy cows for immediate intervention and management in commercial farms.

An Approach for the Evaluation of a Measurement System: A Study on the Use of Machine Learning and Predictions

Quality control during the manufacturing process is an important factor in delivering products in electronics according to planned characteristics and properties. It concerns the capability of the chosen measurement system to perform precise and reliable measurement trials, which is evaluated mainly through the utilization of measurement system analysis. In order to reduce time effort and to partially automate these operations, a methodology for the prediction of a part of the dataset through applying the Neural Net algorithm is proposed in this paper in two scenarios: (1) when two metrology experts are involved in the measurement in three trials and the data of a third specialist are predicted and (2) when three metrology specialists collect data in two trials and the data of the third trial are predicted. The developed predictive models in these two scenarios are assessed and they are characterized by high accuracy. Gage repeatability and reproducibility analysis are used to evaluate the measurement systems based on original and partially artificial datasets as the comparative results outline the suitability of the proposed approach, due to the proximity of the obtained values.

1823-PUB: Expanded Dataset Improves Accuracy of Wrist-Worn Noninvasive BGM Using AI

Background and Aims: To determine accuracy, safety and specificity of a novel non-invasive wrist-worn continuous BGM device which analyses resonance shifts in the microwave spectrum using AI. We present results from an ongoing study in patients with T1D and T2D from an expanded dataset (cohorts 1-3). Methods: In this open, pilot, adaptive design study, subjects (N=5/cohort) attended 4 test occasions (n=2/session), each ≤7 days apart. Devices automatically collected data every 60 secs for 500 msec over 3 hours/session, with plasma glucose measured every 5 mins. A global AI model was evaluated by MARD using venous blood glucose. Interim results from every 5 completed patients informed the next device iteration with 10 iterations possible across the study. Results: Data from each cohort was used to train a neural net algorithm to predict a new trial. Each cohort improved overall MARD prediction accuracy. All analyses followed a leave-one-trial-out methodology where the data for the omitted trial was predicted for each analysis cycle. Using the first cohort, the MARD was 21%, adding the second reduced MARD to 15% and with all 3 cohorts the predictive MARD was 13%. Improvement in the SEG plots showed data falling into low risk SEG categories. Conclusions: We have shown that by giving the neural network an expanded dataset, the MARD decreased to nearer commercially available minimally-invasive BGMs by 38% from a predictive MARD of 21% to 13%. Disclosure M.S.Chaudhry: None. G.J.Dunseath: None. J.Ryan: None. L.Barlow: None. I.C.Carrillo masso: Employee; Afon Technology Ltd. J.H.Crane: None. M.R.A.Qureshi: Employee; Afon Technology Ltd. S.C.Bain: Advisory Panel; Novo Nordisk, Sanofi, Lilly, Boehringer Ingelheim Inc., AstraZeneca. S.D.Luzio: None. C.M.Handy: None. B.Love: Consultant; Afon Technologies LTD. N.M.M.Silva: None. L.M.Ferreira: None. K.Wareham: None.

0275 Deep Learning-based Sleep Detection using Torso Patch Vital Signs Improves Sleep-Wake Detection over Wrist Actigraphy

Abstract Introduction Motion-based wearable sensors, typically on wrist, have long been used for free-living sleep detection and quantification. However, it is hard to differentiate sleep from sedentary awake time by immobility alone. Vital signs, like heart rate and respiration rate, can greatly enhance determination of wake-sleep state, and are easily monitored with newer wearable sensors. Deep learning techniques are particularly adept at learning labeled physiological states. By combining movement plus vital signs in a deep neural network algorithm, improved sleep detection, fragmentation and sleep staging should be possible compared to activity alone. We report on performance of a deep learning sleep detection and REM/NREM algorithm providing 24-hour evaluation with high specificity using data from a torso-wearable patch sensor as compared to polysomnography (PSG). Methods Twenty-six healthy adults (mean age 53.7 years, 81% female) contributed 150 nights of PSG during laboratory visits, during which participants simultaneously wore a multi-day skin-adherent patch with continuous single-lead ECG and 3-axis accelerometer streams, as well as a wrist activity monitor. A pre-trained deep neural network algorithm generated epoch-level Wake/REM/NREM classification (Sleep equals REM plus NREM) using vital signs and movement derived from the patch sensor ECG and accelerometer waveforms and was then compared to expert human staging of PSGs. The wrist actigraphy sleep-wake determinations (Actiware) were also compared to PSG. Results Data includes 900 hours sleeping and 139 hours awake, of which 195 hours of sleep were in REM state. Using patch data, the deep neural net algorithm achieved 92% sensitivity and 85% specificity to detect sleep as compared to PSG; REM was detected with 85% sensitivity and 97% specificity. By comparison, the wrist motion-based algorithm only exhibited 33% specificity and 95% sensitivity, essentially overcalling immobile wake as sleep. Conclusion Sleep evaluation in free-living environments with wearable sensors can be greatly improved over conventional motion-based wrist sensors by leveraging continuous vital signs. Deep learning-trained neural network algorithms are particularly effective for use with such data, as demonstrated with this algorithm. Support (if any) R01 HL140580 and P01 AG011412

Full-scale simulation and experimental verification of the phase-transition temperature of a VO2 nanofilm as smart window materials

Vanadium dioxide (VO2), a highly important thermotropic phase-transition material, undergoes a reversible metal-insulator transition near temperatures of 68 °C and has potential applications in the energy-efficient smart window field. Recently, it has been found that the phase-transition temperature of VO2 films with nanometer thickness approached the ideal of room temperature (30 °C); however, deciphering this nanoscale phase-transition mechanism is still difficult in experiments and computer simulations. Here, a full-scale computational model was carefully constructed using molecular dynamics and first principles based on the artificial neural net algorithm and fully validated via accurate atomic layer deposition experiments. Overall, the machine-learning training results indicated that experimental observations and numerical simulation were in good agreement, demonstrating the feasibility and accuracy of our full-scale nanothermodynamics model. The computational simulations showed that the surface 3-layer atomic occupancy of the VO2 nanomaterials was a critical regulator of the phase transition temperature, implying as an important strategy for phase transition behavior in large-scale VO2 nanofilms designed on first principles. Moreover, the precise atomic layer deposition results agree with the simulation calculations by 98%, revealing that the computational model effectively and accurately combines with the experiments, and thus can accurately regulate the phase transition temperature of the VO2 nanofilm. This work was of crucial importance for potential applications of VO2 film in smart windows.

“Machine micro/nano optics scientist”: Application and development of artificial intelligence in micro/nano optical design

Micro/nano optical materials and devices are the key to many optical fields such as optical communication, optical sensing, biophotonics, laser, and quantum optics, etc. At present, the design of micro/nano optics mainly relies on the numerical methods such as Finite-difference time-domain (FDTD), Finite element method (FEM) and Finite difference method (FDM). These methods bottleneck the current micro/nano optical design because of their dependence on computational resources, low innovation efficiency, and difficulties in obtaining global optimal design. Artificial intelligence (AI) has brought a new paradigm of scientific research: AI for Science, which has been successfully applied to chemistry, materials science, quantum mechanics, and particle physics. In the area of micro/nano design AI has been applied to the design research of chiral materials, power dividers, microstructured optical fibers, photonic crystal fibers, chalcogenide solar cells, plasma waveguides, etc. According to the characteristics of the micro/nano optical design objects, the datasets can be constructed in the form of parameter vectors for complex micro/nano optical designs such as hollow core anti-resonant fibers with multi-layer nested tubes, and in the form of images for simple micro/nano optical designs such as 3dB couplers. The constructed datasets are trained with artificial neural network, deep neural network and convolutional neural net algorithms to fulfill the regression or classification tasks for performance prediction or inverse design of micro/nano optics. The constructed AI models are optimized by adjusting the performance evaluation metrics such as mean square error, mean absolute error, and binary cross entropy. In this paper, the application of AI in micro/nano optics design is reviewed, the application methods of AI in micro/nano optics are summarized, and the difficulties and future development trends of AI in micro/nano optics research are analyzed and prospected.

ROADSIDE FOREST MODELING USING DASHCAM VIDEOS AND CONVOLUTIONAL NEURAL NETS

Abstract. Tree failure is a primary cause of storm-related power outages throughout the United States. Roadside vegetation management is therefore critical to electric utility companies to prevent power outages during extreme weather conditions. It is difficult to execute roadside vegetation management practices, at the landscape level, without proper monitoring of roadside forests’ physical structure and health condition. Remote sensing images and LiDAR are widely used to characterize the forest edge; however, the limitation on the temporal and spatial resolution for most of that dataset is a big challenge. Also, there is a need for a ground-level dataset that provides the vertical profile of the forest trees so that we can more accurately characterize the forest structure and health and recommend the optimal management strategies according to the local forest conditions. For the first time, we introduced Dashcam videos as an alternative to the existing aerial remote sensing data sources to characterize the roadside forest condition using the deep learning (DL) convolutional neural net (CNN) algorithms. In this study, we used dashcam videos taken during the leaf-on and leaf-off conditions and various weather conditions along the roadside. We trained a DLCNN model based on the U-Net and YOLO v5 architectures to classify the multilayer vegetation and detect utility poles and tree trunks alongside the road. Our experiment results suggest that a dashcam can be a viable alternative and complementary way to characterize the roadside vegetation and can be used in the management of roadside forests as a cost-effective data acquisition mechanism for utility companies.

Crime index based on text mining on social media using multi classifier neural-net algorithm

Everyday criminal issues appear on social media, even some crime news is often disturbing to the public but it gives a warning to the public to remain careful and alert to the surrounding environment. However, following large amounts of crime information on social media is not effective, especially for busy people. Therefore, there is a need to efficiently and effectively summarize information in a way that is meaningful and easy to see, attracts people’s attention, and can be used by law enforcement officials. The purpose of this study is to present the index crime based on social media by looking for patterns of crime. This study proposes the projected index crime based on crime trends by using text mining to classify tweet texts and post contents into 10 crime classes. The classification method uses the neural-net multi classifier algorithm which has several classifiers namely logistic regression, naïve bayes, support vector machine (SVM), and decision tree in parallel. In this approach, the classifier that provides the best accuracy will be the winning classifier and will be used in the next learning process. In this experiment, in using the multi classifier neural-net, the logistics regression classifier often provides the best accuracy.

Performance metrics for marine mammal signal detection and classification.

Automatic algorithms for the detection and classification of sound are essential to the analysis of acoustic datasets with long duration. Metrics are needed to assess the performance characteristics of these algorithms. Four metrics for performance evaluation are discussed here: receiver-operating-characteristic (ROC) curves, detection-error-trade-off (DET) curves, precision-recall (PR) curves, and cost curves. These metrics were applied to the generalized power law detector for blue whale D calls [Helble, Ierley, D'Spain, Roch, and Hildebrand (2012). J. Acoust. Soc. Am. 131(4), 2682-2699] and the click-clustering neural-net algorithm for Cuvier's beaked whale echolocation click detection [Frasier, Roch, Soldevilla, Wiggins, Garrison, and Hildebrand (2017). PLoS Comp. Biol. 13(12), e1005823] using data prepared for the 2015 Detection, Classification, Localization and Density Estimation Workshop. Detection class imbalance, particularly the situation of rare occurrence, is common for long-term passive acoustic monitoring datasets and is a factor in the performance of ROC and DET curves with regard to the impact of false positive detections. PR curves overcome this shortcoming when calculated for individual detections and do not rely on the reporting of true negatives. Cost curves provide additional insight on the effective operating range for the detector based on the a priori probability of occurrence. Use of more than a single metric is helpful in understanding the performance of a detection algorithm.

Satellite-based analysis of classification algorithms applied to the riparian zone of the Malaya Kokshaga river

The paper comparatively analyses the accuracy of land cover classification in the riparian zone of the Malaya Kokshaga river in the Mari El Republic of Russia using Sentinel-2A satellite images with the algorithms of supervised classification: Maximum Likelihood (ML), Decision Tree (DT) and Neural Net (NN) in the ENVI-5.2 software package. Six main classes of land cover were identified based on field studies: coniferous, mixed (deciduous), shrublands, herbaceous, and water. The assessment of the area and the structure of land cover showed that forest covers 76% of the entire territory of the riparian area of the Malaya Kokshaga river. The analysis of the results of thematic mapping shows that the overall classification accuracy obtained by the ML algorithm is 96.09%, by NN - 94.51%, and by DT - 86.54%. The producer’s accuracy and user’s accuracy for most classes have the maximum value when the ML algorithm is used. For the NN algorithm, the maximum value of producer’s accuracy is observed for the mixed (deciduous) class, while for the DT algorithm – for the coniferous. When classified using all three algorithms the water and bare land classes were mixed, which requires more detailed work when estimating riparian forest ecosystems.

Machine Learning Analysis for Phenolic Compound Monitoring Using a Mobile Phone-Based ECL Sensor.

Machine learning (ML) can be an appropriate approach to overcoming common problems associated with sensors for low-cost, point-of-care diagnostics, such as non-linearity, multidimensionality, sensor-to-sensor variations, presence of anomalies, and ambiguity in key features. This study proposes a novel approach based on ML algorithms (neural nets, Gaussian Process Regression, among others) to model the electrochemiluminescence (ECL) quenching mechanism of the [Ru(bpy)3]2+/TPrA system by phenolic compounds, thus allowing their detection and quantification. The relationships between the concentration of phenolic compounds and their effect on the ECL intensity and current data measured using a mobile phone-based ECL sensor is investigated. The ML regression tasks with a tri-layer neural net using minimally processed time series data showed better or comparable detection performance compared to the performance using extracted key features without extra preprocessing. Combined multimodal characteristics produced an 80% more enhanced performance with multilayer neural net algorithms than a single feature based-regression analysis. The results demonstrated that the ML could provide a robust analysis framework for sensor data with noises and variability. It demonstrates that ML strategies can play a crucial role in chemical or biosensor data analysis, providing a robust model by maximizing all the obtained information and integrating nonlinearity and sensor-to-sensor variations.

Phytoplankton Bloom Dynamics in the Baltic Sea Using a Consistently Reprocessed Time Series of Multi-Sensor Reflectance and Novel Chlorophyll-a Retrievals

A relevant indicator for the eutrophication status in the Baltic Sea is the Chlorophyll-a concentration (Chl-a). Alas, ocean color remote sensing applications to estimate Chl-a in this brackish basin, characterized by large gradients in salinity and dissolved organic matter, are hampered by its optical complexity and atmospheric correction limits. This study presents Chl-a retrieval improvements for a fully reprocessed multi-sensor time series of remote-sensing reflectances (Rrs) at ~1 km spatial resolution for the Baltic Sea. A new ensemble scheme based on multilayer perceptron neural net (MLP) bio-optical algorithms has been implemented to this end. The study documents that this approach outperforms band-ratio algorithms when compared to in situ datasets, reducing the gross overestimates of Chl-a observed in the literature for this basin. The Rrs and Chl-a time series were then exploited for eutrophication monitoring, providing a quantitative description of spring and summer phytoplankton blooms in the Baltic Sea over 1998–2019. The analysis of the phytoplankton dynamics enabled the identification of the latitudinal variations in the spring bloom phenology across the basin, the early blooming in spring in the last two decades, and the description of the spatiotemporal coverage of summer cyanobacterial blooms in the central and southern Baltic Sea.

Deception Detection in Group Video Conversations using Dynamic Interaction Networks

Predicting groups of people who are jointly deceptive is critical in settings such as sales pitches and negotiations. Past work on deception in videos focuses on detecting single deceivers and uses facial or visual features only. We propose the concept of Face-to Face Interaction Networks (FFINs) and Negative Interaction Networks (NINs) to model interactions within a group of people. The use of FFINs and NINs in this paper enables us to leverage network relations in predicting face-to-face deception for the first time. We will use a dataset of 185 videos from a deception-based game called Resistance for deception based prediction using FFINs. We first characterize the behavior of individuals, pairs, and groups of deceptive participants compared to non-deceptive participants. Our analysis reveals that less engaged deceivers are identified early. Moreover, pairs of deceivers tend to avoid mutual interaction and focus their attention on non-deceivers. In contrast, non-deceivers interact with everyone equally. We propose the notion of Negative Interaction Networks (NINs) and create a belief propagation neural net algorithm called BPNN based on dynamic FFINs and NINs to detect deceivers from videos that are just 1 minute long. We show that our method outperforms recent state-of-the-art computer vision, graph embedding, and ensemble methods by at least 20.9% AUROC in identifying deception from videos.

Digital image analysis of ultrasound images using machine learning to diagnose pediatric nonalcoholic fatty liver disease

PurposePrevalence of nonalcoholic fatty liver disease (NAFLD) in children is rising with the epidemic of childhood obesity. Our objective was to perform digital image analysis (DIA) of ultrasound (US) images of the liver to develop a machine learning (ML) based classification model capable of differentiating NAFLD from healthy liver tissue and compare its performance with pixel intensity-based indices. MethodsDe-identified hepatic US images obtained as part of a cross-sectional study examining pediatric NAFLD prevalence were used to build an image database. Texture features were extracted from a representative region of interest (ROI) selected from US images of subjects with normal liver and subjects with confirmed NAFLD using ImageJ and MAZDA image analysis software. Multiple ML classification algorithms were evaluated. ResultsFour-hundred eighty-four ROIs from images in 93 normal subjects and 260 ROIs from images in 39 subjects with NAFLD with 28 texture features extracted from each ROI were used to develop, train, and internally validate the model. An ensembled ML model comprising Support Vector Machine, Neural Net, and Extreme Gradient Boost algorithms was accurate in differentiating NAFLD from normal when tested in an external validation cohort of 211 ROIs from images in 42 children. The texture-based ML model was also superior in predictive accuracy to ML models developed using the intensity-based indices (hepatic-renal index and the hepatic echo-intensity attenuation index). ConclusionML-based predictive models can accurately classify NAFLD US images from normal liver images with high accuracy using texture analysis features.

A novel method of determining interatomic potential for Al and Al-Li alloys and studying strength of Al-Al3Li interphase using evolutionary algorithms

Aluminum, alloyed with Lithium, is used extensively in aerospace and cryogenic applications due to its high forming ability and desirable mechanical properties. However, a number of short-comings of these alloys led to intense research on the intrinsic phases and their effects. One such metastable phase, δ’ (Al3Li) has immense impact on the properties, requiring extensive probing. Taylor-made interatomic potentials for this system is however unavailable. Therefore, in this study, the interatomic potentials for Al and binary Al-Li have been calculated based on second nearest neighbor modified embedded atom method (2NN MEAM) formulization. The potentials are created based on optimizing parameters for MEAM potential using reference vector guided Genetic Algorithm to predict certain physical properties like cohesive energy, elastic constants, lattice constant, stacking fault energy etc. in reasonable agreement with Density Functional Theory (DFT) calculations. An Evolutionary Deep Neural Net (EvoDN2) algorithm created the objective functions for optimization. The potential predicted for Aluminum can simultaneously produce six to seven various properties with minimum amount of error, thus validating the method. This new MEAM potential is used to study the interphase strength of Al3Li-Al. Metamodels are constructed for this purpose using this new potentials through EvoDN2 algorithm. Applying multi-objective evolutionary algorithms, the metamodels are then utilized to study the optimized working conditions for maximum interphase energy and minimum strain at failure, as this will signify maximum interphase strength.

Evaluation and comparison of a machine learning cloud identification algorithm for the SLSTR in polar regions

A Feed Forward Neural Net (NN) approach to distinguish between clouds and the surface has been applied to the Sea and Land Surface Temperature Radiometer in polar regions. The masking algorithm covers the Arctic, Antarctic and regions typically classified as the cryosphere such as northern hemisphere permafrost. The mask has been trained using collocations with the CALIOP active lidar, which in narrow strips provide more accurate detection of cloud, and was subsequently evaluated as a function of cloud type and surface type. The mask was compared with the existing operational Bayesian and Empirical cloud masks by eye and also statistically using CALIOP data. It was found to perform exceptionally well in the polar regions. The Kuiper skill score improved from 0.28, for the operational Bayesian and 0.17 for the Empirical masks to 0.77 for the NN. The NN algorithm also has a much more homogeneous performance over all surface types. The key improvement came from better identification of clear scenes; for the NN mask, the same performance in terms of contamination of cloudy pixels in the sample of identified clear pixels can be achieved while retaining 40% of the clear pixels compared with 10% for the operational cloud identification. The algorithm performed with almost the same skill over sea and land. The best performance was achieved for opaque clouds while transparent and broken clouds showed slightly reduced accuracy.

Mechanical properties of micro-alloyed steels studied using a evolutionary deep neural network

ABSTRACT The existing Evolutionary Neural Net Algorithm (EvoNN) for data-driven modeling has been augmented during this study using an evolutionary deep neural net strategy to give rise to a novel algorithm named EvoDN, which has been further upgraded to an improved version named EvoDN2. This study reports an application of EvoDN2 to study vanadium and niobium based micro-alloyed steels. For this purpose, a dataset for ultimate tensile strength, elongation and Charpy impact energy at −40°C is collected and trained using aforementioned EvoNN, EvoDN2, and another in house algorithm named Bi-objective genetic programming (BioGP). This trained models are then optimized to get optimized properties using a constrained version Reference Vector Evolutionary Algorithm (cRVEA). The results are thoroughly compared with the existing correlations and prior work and found to be well within the acceptable range.

Forecasting episodes of atrial fibrillation using a new machine learning algorithm

The performance of preventive pacing algorithms (PPA) that can reduce atrial fibrillation (AF) burden in cardiac electronic implanted devices (CEID's) has been disappointing. This can be due to the fact that the time frame between the start of AF and the implementation of preventive stimulation and anti-tachycardia pacing algorithms is too short. Nowadays, machine learning algorithms are powerful tools and have proven their ability to anticipate events. Therefore, we assessed forecasting of paroxysmal AF minutes before its onset using machine learning techniques. To forecast an oncoming AF episode in individual patients using machine learning techniques and to assess when a change in heart rate indicates the beginning of a possible forecast. We analysed retrospectively the raw data of a data set of 10484 ECG Holter monitorings searching for AF episodes of more than 30 sec. A hundred and forty patients presented paroxysmal AF (286 recorded AF episodes). We used only RR intervals to predict the presence of AF. It should be emphasized that we did not integrate clinical variables. We developed a deep and recurrent neural net machine learning algorithm. This algorithm was trained on using 90% of patients records and tested on the remaining 10%. Rather than attempting to predict the exact RR-interval at which AF would or would not happen our algorithm attempted to predict if AF would happen within windows of time of 30 RR intervals ( Table 1 ). The performance figures are summarized in the table. Based upon this retrospective study, AF can be forecasted thirty beats before its onset with a reasonable sensitivity, using our machine learning algorithm. We believe that its embedding in CEID's could lower AF burden by improving results of PPA in selected patients.