Inappropriate Detection Research Articles

Rhythm classifier: artificial intelligence for rhythm classification in insertable cardiac monitor

Abstract Background Though detections improved significantly, separating atrial tachycardia/atrial flutter (AT/AFL) from atrial fibrillations (AF) in insertable cardiac monitors (ICM) remains a challenge. Objective Using artificial intelligence (AI) algorithms to classify AF and AT/AFL in ICM detected episodes, which may reduce episode review burden. Methods Both raw EGM signal and the QRS diminished version of the raw signal were used to extract 14 AF and AT/AFL features from each episode. Extracted features were then represented as an ensemble of features and used to train and test the AI model. Modified ResNet18 network was used as the AI model. All episodes from a random selection of 80% of patients were used as training dataset, another 10% were used as validation, and 10% were used as test dataset, to evaluate model performance. AI model probability threshold was chosen based on validation set results. Episode detection sensitivity, specificity, and area under the curve (AUC) of receiver operating characteristics (ROC) curve in the independent test dataset are reported. Results AF and AT/AFL episodes detected in 918 LINQ ICM patients (5,253 AF and 4,705 AT/AFL – 745 pts and 8,950 false detection episodes) were included. Training dataset consisted of 15,183 episodes (735 pts), validation set consisted of 1,901 episodes (91 pts), and test set consisted of 1,824 episodes (92 pts). At chosen threshold in validation set, AF and AT/AFL detection had sensitivity of 90% and 73.3%, respectively, and specificity of 93.2% and 95%, respectively, with AUC of 0.966. In test data, AF and AT/AFL obtained relative sensitivity of 84.7% and 85.7%, respectively, with specificity of 88% and 86.5%, respectively (Fig. A). With ROC-AUC of 0.956 (Fig. B), the relative diagnostic yield of AF and AT/AFL on test data was 95.6%. Conclusion The trained AI model provides a relative sensitivity of about 85% for AF vs. AT/AFL classification and reduces inappropriate ICM detections by 79%. Adding more episodes may improve the performance.

Feasibility of using chest strap and dry electrode system for longer term cardiac arrhythmia monitoring: Results from a pilot observational study

Background and aimCardiac arrhythmia diagnostic yield improves with increased duration of monitoring. We investigated patient comfort, diagnostic quality of ECG, and arrhythmia diagnostic yield using a single lead longer term external cardiac monitor (ECM). MethodsThe observational ECM feasibility study enrolled patients with increased risk of cardiac arrhythmia. The ECM investigational prototype was designed using a chest strap with dry electrodes connected to module capable of triggered loop recording of ECG, and automatic detection of arrhythmia. In group-A of study (24-h inpatient), patients wore ECM and Holter that recorded ECG from the ECM and adhesive electrodes. In group-B of study (12-weeks ambulatory), at monthly follow-ups patients filled out a comfort survey and device stored arrhythmia episodes were reviewed. ResultsThe study enrolled 34 patients (38 % females, average age 57.5 years, 65 % had palpitations, 12 % had syncope). Diagnostic quality ECG was recorded on 76.5 % of the monitoring duration in 12 of 20 patients with reviewable data in group-A, with motion artifacts causing loss in ECG signal for 18.7 % of the time. In 14 patients in group-B, 94.9 % of the survey responses indicated that ECM was comfortable to wear. Cardiac arrhythmia was observed in 4 of 17 patients (24 %) in group-A and 9 of 14 patients (64 %) in group-B in device recorded episodes. All ECM detected pause and tachycardia were inappropriate detections due to motion artifacts and temporary device removal. ConclusionThe chest strap-based ECM device was mostly comfortable to wear and recorded diagnostic quality ECG in three-fourth of monitoring period. Cardiac arrhythmia was observed in 64 % of patients over 3-month monitoring along with large number of motion artifact induced inappropriate detections.

Feasibility, safety, and short-term follow-up of defibrillator lead at left bundle branch area pacing location: A pilot study

BackgroundLeft bundle branch area pacing (LBBAP) has been increasingly adopted as an alternative modality to cardiac resynchronization therapy (CRT). The feasibility and safety of using an LBBAP lead to provide sensing of ventricular arrhythmia in patients receiving an implantable cardioverter-defibrillator (ICD) with CRT has been demonstrated recently. ObjectivesThe purpose of our study was to analyze the feasibility, safety, and short-term follow-up of a traditional defibrillator lead at the LBBAP location. MethodsPatients who underwent successful LBBAP defibrillator using DF-1/DF-4 lead and delivery catheter were included in the study. Defibrillation threshold (DFT) testing was performed after implantation to assess the ability of the LBBAP defibrillator lead to sense and provide appropriate therapy for ventricular arrhythmia. ResultsAlthough the ICD lead could be successfully deployed in the left bundle branch area in 7 of 8 patients, it was repositioned to the right ventricular (RV) apex because of atrial oversensing in 1 patient and cheesy septum in another patient. Acute procedural success was 62.5% (5/8 patients). Mean patient age was 62.6 ± 21.6 years. Mean procedural duration was 115.6 ± 38.1 minutes, with LBBAP defibrillator lead fluoroscopy duration of 10.6 ± 3.5 minutes. Mean capture threshold was 0.58 ± 0.23V/0.4 ms, sensed R-wave amplitude 9.6 ± 2.2 mV, pacing impedance 560 ± 145 Ω, and shock impedance 65.4 ± 5.5 Ω. Defibrillation testing was successful in inducing ventricular fibrillation and could be sensed and reverted promptly by the shock delivered through the lead. During mean follow-up of 3.8 ± 2.2 months, pacing parameters remained stable. No episodes of inappropriate arrhythmia detection or therapy delivery occurred during follow-up. ConclusionLBBAP defibrillator is feasible, safe, and effective during short-term follow-up. DFT testing at the time of implantation will help to ensure appropriate sensing and treatment of ventricular arrhythmias.

First human safety and effectiveness study of defibrillation with a novel patch wearable cardioverter-defibrillator.

Wearable cardioverter-defibrillators (WCDs) are indicated in patients at risk of sudden cardiac arrest who are not immediate candidates for implantable defibrillator therapy. Limitations of existing WCDs include poor compliance and high false alarm rates. The Jewel is a novel patch-WCD (P-WCD) that addresses these limitations with an adhesive-based design for near-continuous wear and a machine learning algorithm designed to minimize inappropriate detections. This was a first-in-human study of the Jewel P-WCD conducted in an electrophysiology (EP) lab to determine the safety and effectiveness of the device in terminating ventricular tachycardia/ventricular fibrillation (VT/VF) with a single shock. The aim was to evaluate the safety and effectiveness of terminating VT/VF with a single shock using the Jewel P-WCD. This was a first-in-human, prospective, single-arm, single-centre study in patients scheduled for an EP procedure in which VT/VF was expected to either spontaneously occur or be induced. The Jewel P-WCD was placed on consented patients; upon confirmation of VT/VF, a single shock (150 J) was delivered via the device. A group sequential design and Pocock alpha spending function was used to measure the observed proportion of successful VT/VF single-shock terminations. The endpoint was achieved if the lower confidence limit exceeded the performance goal of 62%, using a one-sided lower 97.4% exact confidence bound. Of 18 eligible subjects, 16 (88.9%, 97.4% confidence bound: 65.4%) were successfully defibrillated with a single shock, exceeding the primary endpoint performance goal with no adverse events. This first-in-human evaluation of the Jewel P-WCD demonstrated the safety and effectiveness of terminating VT/VF. URL: https://clinicaltrials.gov/; Unique identifier: NCT05490459.

Progressive Feature Extraction with a Greedy Network-growing Algorithm

In this paper, a new information theoretic method called the greedy network-growing algorithm is proposed. The method is called "greedy," because a network with this algorithm grows while absorbing as much information as possible from outside. The method is based upon information theoretic competitive learning and can solve the fundamental problems inherent in competitive learning, such as the dead neurons and inappropriate number of neurons problems. The new model can grow networks by repeatedly maximizing information content and by gradually extracting salient features from input patterns. Because the new model can cope with inappropriate feature detection in the early stage of learning, extracted features should cover most of the input patterns. The new method is applied to political data analysis, medical data analysis, and information science education analysis. Experimental results confirm that the new method can acquire significant information and that more explicit features can be extracted.

High incidence of inappropriate alarms due to artifact sensing and misclassification of episodes in patients with wearable cardioverter-defibrillators

Abstract Background The wearable cardioverter-defibrillator (WCD) has become a useful tool used for temporary protection from sudden cardiac death. However, since the WCD uses surface electrodes to detect arrhythmias, it is prone to inappropriate detection. Although shock conversion rates for the WCD are high for detected events, its efficacy in clinical practice tends to be degraded by patient noncompliance. Reasons include wear discomfort and frequent false alarms, that may interrupt sleep and generate anxiety. Data on the incidence of false alarms emitted by the WCD and their predictors are rare. Purpose The aim of our study was to assess both artifact and misclassification burden and wear compliance in patients with a WCD. Methods and Results We conducted a single-centre retrospective observational study analyzing patients with a WCD prescribed at our institution. In total, in 134 patients (mean age 51.7 ± 13.8 ys, 79.1% male) provided with a WCD, arrhythmia recordings were analyzed and categorized as either non-sustained ventricular tachycardia, sustained ventricular tachycardia / fibrillation, and artifact sensing or misclassified episodes. Indication for WCD prescription was both primary and secondary prophylactic. All patients received optimal medical therapy. Baseline evaluation in all patients included transthoracic echocardiography, 12-lead electrocardiogram, and baseline laboratory values. A total of 3019 false WCD alarms were documented in 134 patients (average number of false alarms 22.5 ± 130.4 episodes per patient) over a mean WCD wearing time of 71.5 ± 70.9 days (daily wearing time 20.2 ± 5.9 hs). In total, in 78 patients (58.2%), either artifact sensing (41.8%), misclassified episodes (3.7%) or both (9.7%) occurred. Misclassified episodes included sinus tachycardias, atrial flutter, atrial fibrillation, ventricular extrasystoles, and intermittent bundle branch block. At multivariate analysis, WCD wearing time (HR 1.02; 95% CI 1.01-1.02; P < 0.001) and loop diuretics (HR 0.16; 95% CI 0.06-0.44; P < 0.001) were identified as independent predictors of artifact episode occurrence. In terms of the frequency of artifact episodes, loop diuretics (HR -0.11; 95% CI interval -0.21-0.0001; P = 0.049), angiotensin receptor-neprilysin inhibitors (ARNIs) (HR -0.11; 95% CI -0.22-0.01; P = 0.03), and the R-amplitude (HR -0.17; 95% CI -0.27-0.07; P = 0.01) of the WCD baseline electrocardiogram (ECG) were independent predictors. In addition, WCD wearing time (HR 1.00; 95% CI 1.00-1.01; P = 0.04), atrial fibrillation (HR 4.03; 95% CI 1.15-14.11; P < 0.001), and ARNIs (HR 3.17; 95% CI 1.05-9.52; P < 0.001) were significantly associated with the occurrence of misdetected / misclassified episodes. Conclusions In conclusion, In WCD patients with a ZOLL LifeVest system, false alarms emitted by the device were frequent and most common caused by artefacts.

Extravascular Implantable Cardioverter-Defibrillator Sensing and Detection in a Large Global Population

BackgroundThe extravascular (EV) implantable cardioverter-defibrillator (ICD) includes features to address sensing and arrhythmia detection challenges presented by its substernal lead location. ObjectivesIn this study, the authors sought to evaluate sensing and detection performance in 299 patients discharged with an EV-ICD in the global pivotal study. MethodsWe reviewed and adjudicated all induced ventricular fibrillation (VF) episodes and spontaneous device-stored episodes that satisfied rate and duration criteria in a programmed ventricular tachycardia (VT)/VF therapy zone. ResultsAt implantation, all EV-ICDs detected induced VF at the programmed sensitivity; 95.9% detected VF with a 3× safety margin. In follow-up, EV-ICDs detected all 59 VT/VF episodes that sustained until therapy. Of 1,034 non-VT/VF episodes, oversensing caused 87.9% and supraventricular tachycardia caused 12.1%. Therapy was withheld in 80.9%, aborted in 10.6%, and delivered in 8.5%. The most common causes of oversensing were myopotentials (61.2%) and P-wave oversensing (PWOS) (19.9%). Inappropriate shocks occurred in only 3.2% of myopotential episodes, but in 21.8% of PWOS episodes. Myopotential oversensing was more common with Ring-Can sensing (P < 0.0001) and correlated with low R-wave amplitude (P < 0.0001). PWOS occurred almost exclusively with Ring1-Ring2 sensing (P = 0.0001) and began with transient decrease in R-wave or increase in P-wave amplitude (P < 0.0001). In software emulation, a new PWOS discriminator significantly reduced total inappropriate detections. ConclusionsIn a global population, EV-ICD detected induced and spontaneous VT/VF accurately. Although discriminators withheld detection from most non-VT/VF episodes, inappropriate shocks were common. The most common cause was PWOS, which may be reduced by optimizing sensing at implantation and incorporation of the PWOS discriminator, which is now in the current device. (Extravascular ICD Pivotal Study [EV ICD]; NCT04060680)

Inappropriate YouTube content detection and classification by using proposed novel auto-determined k-means clustering and PDBRNN architecture

To ensure a safe and pleasant user experience while watching content on YouTube, it is necessary to identify and classify inappropriate content, especially content that is inappropriate for children. In this work, we have concentrated on establishing an efficient system for detecting inappropriate content on YouTube. Most of the work focuses on manual pre-processing; however, it takes too much time, requires manpower support, and is not ideal for solving real-time problems. To address this challenge, we have proposed an automatic preprocessing scheme for selecting appropriate frames and removing unwanted frames such as noise and duplicate frames. For this purpose, we have utilized the proposed novel auto-determined k-means (PADK-means) algorithm. Our PADK-means algorithm automatically determines the optimal cluster count instead of manual specifications. By doing this, we have solved the manual cluster count specification problem in the traditional k-means clustering algorithm. On the other hand, to improve the system’s performance, we utilized the Proposed Feature Extraction (PFE) method, which includes two pre-trained models DenseNet121 and Inception V3 are utilized to extract local and global features from the frame. Finally, we employ a proposed double-branch recurrent network (PDBRNN) architecture, which includes bi-LSTM and GRU, to classify the video as appropriate or inappropriate. Our proposed automatic preprocessing mechanism, proposed feature extraction method, and proposed double-branch RNN classifier yielded an impressive accuracy of 97.9%.

Analysis of electrotherapy in patients with implantable cardioverter-defibrillator for primary prevention of sudden cardiac death according to remote monitoring data

Aim: To determine the factors associated with development of justified and inappropriate implantable cardioverter-defibrillator (ICD) activations in patients at high risk of sudden cardiac death based on the remote monitoring systemMaterial and Methods. We analyzed remote monitoring data from 2014 to 2022 in 132 patients with ICD for primary prevention of sudden cardiac death (SCD). Patients were divided into 2 groups 1 – patients with persistent paroxysmal tachyarrhythmias; 2 – patients without persistent paroxysms of ventricular tachycardia (VT) / ventricular fibrillation (VF). In case of inappropriate shocks, the causes and possible predictors of their development were analyzed.Results. Of 132 patients, 62 (46.9%) patients appeared to have persistent VT/VF. It was found that the probability of detecting persistent paroxysms of VT/VF decreased with a history of revascularization (p = 0.030) and increased in the absence of amiodarone therapy (p = 0.012), with increasing age (p = 0.035), with decreasing left ventricular ejection fraction (LVEF) less than 35% (p = 0.016). 71 arrhythmic episodes (17.9%) in 27 (20.4%) patients were considered as false detection of tachyarrhythmias. Analyzing the causes of inappropriate electrotherapy, different types of supraventricular tachyarrhythmias in the zone of VT detection (85.9%), including atrial fibrillation (25.4%), in smaller percentage of cases there was discovered T-wave detection 4.2%, noise on the electrode – 2.8%, 7.1% – double counter due to the operation of the cardiac contractility modulation device. When analyzing factors associated with the development of false detection, reliable results were obtained with regard to the presence of a history of atrial fibrillation (p = 0.036), implanted single-chamber ICD (p = 0.028).Conclusion. The development of persistent ventricular tachyarrhythmias was noted in 47% of patients with ICD as a part of primary prevention of sudden cardiac death, and 20.4% had the development of inappropriate detection and electrotherapy. Predictors of their occurrence have been identified, which can be used as development of strategies for shock minimization.

ASSESSMENT OF EFFICIENCY OF ARTIFICIAL INSEMINATION SERVICE IN CATTLE PRODUCTION IN THE WESTERN ZONE OF TIGRAY REGION, ETHIOPIA

The study was conducted before the ignition (before October 2020) of the war in Tigray Region, Ethiopia. Artificial insemination (AI) is the first generation biotechnology and is one of the assisted reproductive techniques (ART) that greatly plays for a faster genetic improvement of farm animals. Though AI was implemented in cattle crossbreeding for many decades in Ethiopia, there is quite insignificant national output of crossbred cattle population (1.54 %). The aim of the survey was to assess the efficiency of AI service in cattle production in the Western Zone of Tigray Region, Ethiopia. Twenty (20) Kebeles and 353 small-scale farmers were purposively involved in the household face-to-face survey interview. The data were analyzed using Statistical Package for Social Sciences (SPSS), and summarized by mean, frequency and percentages. The survey indicated that the existing reproductive efficiency of cattle AI service in the Western Zone of Tigray Region, Ethiopia was extremely poor. The mean number of services per conception (NSC) was very high that could be due to mainly inappropriate heat detections. The recommended, acceptable and excellent grade of NSC fall in the range of 1 − 1.7. The overall conception rate at first insemination (CRFI) was also very poor (20.4 %) and thus calving rate (CR) was poor (20.5 %). About one-third (33.4 %) of the respondents indicated that cows and heifers exhibited heat in 0:01 − 6:00 am of the day. About 30.9 % and 28.6 % of the respondents reported that their cows and heifers were inseminated in 9 − 12 hours and 4 − 8 hours after the onset of heat, respectively. The majority (70.8 %) of the respondents did not practice controlled mating and breeding, and the first preference of 50.1 % of the respondents was natural mating. The most common source of bulls for about 34.3 % of the respondents was from neighbors and followed by own bred bulls (30.0 %). Cattle AI feasibility study and identification of AI strategic interventions should be the first focus to improve the existing poor AI efficiency of the study area.

Standard and extended electrocardiographic screening for the subcutaneous implantable cardioverter-defibrillator

Abstract Funding Acknowledgements Type of funding sources: None. Background The subcutaneous implantable cardioverter-defibrillator (SICD) is effective therapy for sudden cardiac death prevention. The condition for implantation is an electrocardiographic screening test before procedure (ECG-s). Patients who don’t pass are regarded to be at risk of inappropriate detection of tachyarrhythmias with inappropriate interventions. A patient is considered suitable for implant if at least one surface ECG lead (sense vector) is acceptable for all tested postures. Objective The aim of the study was to identify the incidence of failure, the predicting factors and clinical importance of extended ECG-s with right sided lead position and lying on the left and right side of the body. Methods ECG-s to determine suitability for SICD is based and performed using a manufacturer –specific measurement tool. We performed this test in 75 consecutive pts with indication for SICD therapy on the basis of standard manufacturer protocol(lead at the left sternum border: sitting, standing and lying position), and extended for right sternal lead location and two additional body position: lying on the left and right side (10 position for each individual). We analyzed: age, gender, body mass index(BMI), heart rate, heart rhythm(sinus/atrial fibrillation), QRS duration, PR interval, corrected QT interval(QTc), echocardiography parameters, etiology. Results ECG-s failure occurred in 6/75(8%) of patients. Screening results: 1 vector failure rate with the standard device and lead positions: primary(P) 12%, seconadary(S) 17%, alternative(A) 40%, P vs. A and S vs. A p&amp;lt;0,001. Testing with an electrode positioned on the right sternum border- 1 vector failure: (P) 10%,seconadary(S) 17%, alternative(A) 48%, P vs. A and S vs. A p&amp;lt;0,001. In case of failure on left side 50%, 14% and 6% of pts passed the test with the right sided lead position for P, S and A vector respectively, p&amp;lt;0.001. Identified factors predicting ESG-s passing vs. failure included: QTc interval 445,32±35,4 vs. 480,6±37,6 ms for any vector(p=0,034), BMI:26,7±4,7 vs. 29,9±4,7 for vector A( p=0,008 ), EF: 43,3±20,1 vs. 34,0±16,2 for vector A( p=0,027). Sense vector failure rate with the left sternal lead location lying on the left and right side of the body was concordant with the rate lying on the back, p=0,05. For the right parasternal lead location the rate failure was lower lying on the right side for all sense vectors: P, S and A(p=0,045; 0,083 and 0,020) respectively. Conclusions ECG screening failure rate for SICD is comparable with previous reports. The right sternal lead location screening is effective and enables important clinical findings before implantation and in case of lead reposition( defibrillation threshold test failure) during procedure. We recommend to screen routinely on both sternum borders. Some additional body position testing: lying on the left and right side of the body, mimicking sleeping/rest could be an important element to determine eligibility for SICD.

PO-03-050 INCIDENCE OF INAPPROPRIATE ATRIAL FIBRILLATION DETECTION WITH REMOTE CARDIAC MONITORING BY INSERTABLE CARDIAC MONITOR

An increasing number of patients are receiving long-term cardiac monitoring to detect and manage atrial fibrillation (AF). The Reveal LINQ insertable cardiac monitor (ICM) (Medtronic, Minneapolis, MN, U.S.A.), is a miniaturized, leadless, subcutaneous, long-term cardiac monitoring that is widely used to manage arrhythmias. To study the inaccurate (false-positive (FP)) detection of AF during remote cardiac monitoring with the Reveal LINQ ICM. In this non-randomized retrospective study, all clinic patients with a Reveal LINQ ICM implanted between March 2014 and January 2021 were included. The record of the last reported AF for each patient was reviewed. Of 294 consecutive patients, 127 patients (average age= 67.8, Female = 52 (40.9%)) had at least one report of AF. The main reason for implanting cardiac monitoring was AF management (38.6%). There were 34 true AF and 93 inappropriate AF detections. The percentage of FP-reported AF was 73.2%. There was not a significant association between the rate of FP and reasons for cardiac monitoring (P-value = 0.516 by Fisher's exact test). Our findings based on the reason for monitoring are summarized in Table 1. The true diagnosis of reported inappropriate AF were: premature atrial contractions (44.9%) , premature ventricular contractions (11.8%), atrial flutter ( 10.2%) , atrial tachycardia ( 0.8%), premature atrial contractions / ventricular tachycardia (0.8%), noise (4.7%). There is a high rate of inappropriate AF detection by the Reveal LINQ ICM, requiring reviewing the tracing by an expert. More robust algorithms need to be developed for the Reveal LINQ ICM to reduce the rate of FP.

PO-03-055 INCIDENCE OF INAPPROPRIATE PAUSE DETECTION WITH REMOTE CARDIAC MONITORING BY INSERTABLE CARDIAC MONITOR, A REAL-WORLD STUDY

An increasing number of people receive long-term cardiac monitoring with implantable devices due to the growing prevalence of cardiac arrhythmias. The LINQ II is Medtronic’s latest generation insertable cardiac monitor (ICM) (Medtronic, Minneapolis, MN, U.S.A.) Compared to its previous generation, Reveal LINQ ICM, the LINQ II ICM employs artificial intelligence (AI) to reduce the false positive (FP) detection of pause. We previously showed that the rate of FP pause among our patients with Reveal LINQ ICM was 52%, but some researchers showed this rate could be as high as 76%. To detect the incidence of inappropriate (FP) diagnoses of pause during remote cardiac monitoring with the LINQ II ICM. This is a non-randomized retrospective study. All our clinic patients with LINQ II ICM between May 2021 and November 2022 were included. We excluded the patients with incomplete data. The record of all pauses for each patient was reviewed. Of 73 consecutive patients (average age 64.9, Female = 35 (48%)), 10 patients had at least one report of pause, and the total number of reported pauses was 43 (average 4.3 per patient, min = 1, max = 12). There were 15 true pause episodes and 28 inappropriate pause detections. The device’s accuracy in detecting a true pause was 35%. The rate of reported FP pauses was 65%. Five patients had at least one episode of true pause. The specificity of LINQ II ICM to find patients with pause was 50%. Despite using AI, there is a high rate of inappropriate pause detection by the LINQ II ICM, and tracing examination by an expert is essential to confirm the validity of the device report. AI didn’t improve the accuracy of pause detection in our patients. More robust algorithms are required to be developed for the LINQ II ICM to reduce the rate of FP.

A Semi-Supervised Speech Deception Detection Algorithm Combining Acoustic Statistical Features and Time-Frequency Two-Dimensional Features.

Human lying is influenced by cognitive neural mechanisms in the brain, and conducting research on lie detection in speech can help to reveal the cognitive mechanisms of the human brain. Inappropriate deception detection features can easily lead to dimension disaster and make the generalization ability of the widely used semi-supervised speech deception detection model worse. Because of this, this paper proposes a semi-supervised speech deception detection algorithm combining acoustic statistical features and time-frequency two-dimensional features. Firstly, a hybrid semi-supervised neural network based on a semi-supervised autoencoder network (AE) and a mean-teacher network is established. Secondly, the static artificial statistical features are input into the semi-supervised AE to extract more robust advanced features, and the three-dimensional (3D) mel-spectrum features are input into the mean-teacher network to obtain features rich in time-frequency two-dimensional information. Finally, a consistency regularization method is introduced after feature fusion, effectively reducing the occurrence of over-fitting and improving the generalization ability of the model. This paper carries out experiments on the self-built corpus for deception detection. The experimental results show that the highest recognition accuracy of the algorithm proposed in this paper is 68.62% which is 1.2% higher than the baseline system and effectively improves the detection accuracy.

High-power chargers for electric vehicles: are they safe for patients with pacemakers and defibrillators?

Battery electric vehicle (BEV) sales and use are rapidly expanding. Battery electric vehicles, along with their charging stations, are a potential source of electromagnetic interference (EMI) for patients with cardiac implantable electronic devices (CIEDs). The new 'high-power' charging stations have the potential to create strong electromagnetic fields and induce EMI in CIEDs, and their safety has not been evaluated. A total of 130 CIED patients performed 561 charges of four BEVs and a test vehicle (350 kW charge capacity) using high-power charging stations under continuous 6-lead electrocardiogram monitoring. The charging cable was placed directly over the CIED, and devices were programmed to maximize the chance of EMI detection. Cardiac implantable electronic devices were re-interrogated after patients charged all BEVs and the test vehicle for evidence of EMI. There were no incidences of EMI, specifically no over-sensing, pacing inhibition, inappropriate tachycardia detection, mode switching, or spontaneous reprogramming. The risk of EMI on a patient-based analysis is 0/130 [95% confidence interval (CI) 0%-2%], and the risk of EMI on a charge-based analysis is 0/561 (95% CI 0%-0.6%). The effective magnetic field along the charging cable was 38.65 µT and at the charging station was 77.9 µT. The use of electric cars with high-power chargers by patients with cardiac devices appears to be safe with no evidence of clinically relevant EMI. Reasonable caution, by minimizing the time spent in close proximity with the charging cables, is still advised as the occurrence of very rare events cannot be excluded from our results.

Deep learning based whale detection from satellite imagery

In spite of their fascinating and vulnerable status, the number of whales in the world's oceans remain debatable because of their living environment. As per statics, with the efforts made towards their conservation, there have been signs of recovery in the population of whales but still, many remote areas remain unsurveyed and thus the number of whales remain ambiguous. To maintain the balance of ecology it is needed to preserve the whales to balance the ecosystem and a key component of that will be to have a count of whales.To date, whale detection has been carried out using ship/aerial surveys, acoustic surveys, etc. but with the development in the field of Very High Resolution (VHR) satellite imagery, satellite images along with deep learning is also being effectively used for the detection of whales.Most of the previous work carried out for the detection of whales via satellite imagery is based upon the manual detection of whales by observers or scientists and that suffers from a high risk of false and inappropriate detections. There is also limited availability of satellite data in this field and there is a lack of usage of Artificial Intelligence for the automatic detection of whales.In this work we propose, a novel deep learning-based automatic whale detection approach from VHR satellite feed. We created a dataset of satellite images having evident chances of whales or probable whales by combining images from two different sources, used Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) to improve the quality of the images and maintain the natural texture of the images, augmented the captured satellite images to train the model, labelled the dataset as whale and probable whale. Finally, a whale dataset of 6000 whale satellite images is created which is the largest dataset available in this field. A deep learning-based whale detection approach is proposed to detect the existence of whales in the satellite image. The proposed approach performed better in terms of precision, recall, and mAP (Mean Average Precision). The precision achieved by the proposed system is 0.95, recall is 0.85, and mAP is 0.74 which is better as compared to the state-of-the-art work in this field.

An ensemble of features based deep learning neural network for reduction of inappropriate atrial fibrillation detection in implantable cardiac monitors.

Multiple studies have reported on classification of raw electrocardiograms (ECGs) using convolutional neural networks (CNNs). We investigated an application-specific CNN using a custom ensemble of features designed based on characteristics of the ECG during atrial fibrillation (AF) to reduce inappropriate AF detections in implantable cardiac monitors (ICMs). An ensemble of features was developed and combined to form an input signal for the CNN. The features were based on the morphological characteristics of AF, incoherence of RR intervals, and the fact that AF begets more AF. A custom CNN model and the RESNET18 model were trained using ICM-detected AF episodes that were adjudicated to be true AF or false detections. The trained models were evaluated using a test dataset from independent patients. The training and validation datasets consisted of 31,757 AF episodes (2516 patients) and 28,506 false episodes (2126 patients). The validation set (20% randomly chosen episodes of each type) had an area under the curve of 0.996 for custom CNN (0.993 for RESNET18). Thresholds were chosen to obtain a relative sensitivity and specificity of 99.2% and 92.8%, respectively (99.2% and 87.9% for RESNET18, respectively). The performance in the independent test set (4546 AF episodes from 418 patients; 5384 false episodes from 605 patients) showed an area under the curve of 0.993 (0.991 for RESNET18) and relative sensitivity and specificity of 98.7% and 91.4%, respectively, at chosen thresholds (98.9% and 88.2% for RESNET18, respectively). An ensemble of features-based CNNs was developed that reduced inappropriate AF detection in ICMs by over 90% while preserving sensitivity.

ZQBA: A Zero-Query, Boosted Ambush Adversarial Attack on Image Retrieval

Content Based Image Retrieval (CBIR) Systems have been employed in a wide variety of critical applications such as intellectual property management [47], facial recognition [46], and inappropriate content detection [48]. Most CBIRs are vulnerable to adversarial attacks, where small, imperceptible perturbations to input images cause system failure. In this paper, we propose a zero-query, black-box adversarial attack method that simulates an attack setting where the attacker has no knowledge about the CBIR model architecture and is unable to make multiple queries. The proposed method uses an ensemblebased approach, generating one perturbation for an input image that severely hinders the ability of six different CBIR models. Our approach is successfully able to disrupt the relevance of our target image retrieval models with a 65% decrease in Mean Average Precision (mAP) as compared to state-of-the-art UAP [18]. We hope our method serves as a baseline for the evaluation of robustness for future image retrieval research.

PO-620-07 AN ENSEMBLE OF FEATURES BASED DEEP LEARNING NEURAL NETWORK FOR REDUCTION OF INAPPROPRIATE ATRIAL FIBRILLATION DETECTION IN IMPLANTABLE CARDIAC MONITORS

Multiple studies have been reported for electrocardiogram (ECG) classification using deep learning neural networks. Most of these studies use the ECG as input to a convolution neural network (CNN) which then automatically generate features from the raw ECG signal. To develop an application specific deep learning CNNs using custom ensemble of features to reduce inappropriate AF detections in implantable cardiac monitors (ICM). The ensemble of features was developed and combined to form an input signal for deep learning CNN. The ensemble of features were designed based on the electro-physiological characteristics during AF: presence of fibrillation/flutter waves or absence of single p-wave between two R-waves, incoherence of RR intervals, and AF begetting more AF. A small custom CNN model using 6 convolution layers and the publicly available RESNET18 model were used. The deep learning models were trained and validated using more than 60K ICM detected AF episodes that were adjudicated to be true AF or false detections. The trained models were evaluated using an independent test dataset of ICM detected and adjudicated AF episodes from patients who were not included in the training and validation dataset. The training and validation dataset consisted of 31,768 true AF episodes (2516 patients) and 28,527 false episodes (2126 patients). The validation set used 20% of randomly chosen episodes of each type. The independent test set consisted of 4546 true AF episodes (418 patients) and 5384 false episodes (605 patients). In the validation set, the custom CNN had an area under ROC curve (AUC) of 0.996 (0.993 for RESNET18) and a threshold for discrimination was defined such that a relative sensitivity and specificity of 99.2% and 92.8% (99.2% and 87.9% for RESNET18) was obtained. The performance results for the two models in the independent test dataset for the pre-defined threshold is shown in the figure, with relative sensitivity and specificity of around 99% and 90% respectively. An ensemble of features based deep learning CNN was developed that reduced inappropriate AF detection in ICM by over 90% while preserving sensitivity for detection of true AF. For an application specific feature based CNN, a smaller sized custom CNN performed as well as a larger sized established CNN.

Automated asphalt pavement damage rate detection based on optimized GA-CNN

Pavement needs to be maintained from the moment its service life begins. The maintenance strategy is mainly based on pavement quality indexes, such as the road damage rate (DR). Accurate pavement distress detection is generally costly and complicated. An inappropriate pavement distress detection strategy could yield a low efficiency of budget usage and untreated road diseases. This study describes an innovative vision-based pavement crack detection strategy that provides a direct pavement surface condition index (PCI) for a specific pavement location. This strategy was achieved with the application of the convolutional neural network (CNN) algorithm to mine a database that contains over five thousand pavement distress images to classify the pavement crack type. To improve the accuracy of the network, a genetic algorithm (GA) was employed to optimize the model. To simultaneously consider the efficiency and accuracy, an enhanced image processing method was used to measure the DR of the pavement. The model input image size is 100 × 100 pixels and the test results showed that the proposed method has satisfactory performance in robustness, accuracy, and processing speed. The accuracy of classifying different crack types reached 98%, and the processing time for an image is 0.047 s. This research, for the first time, proposed an ellipse fitting method to calculate the block of mesh cracks. The overall accuracy in detecting the pavement damage rate reached 90%. This study demonstrates the potential of an innovative deep learning method in automated pavement quality index calculations to provide direct guidance for long-term asphalt pavement optimal rehabilitation and maintenance (R&M) decisions.