On-device artificial intelligence: mobile solution for detecting severe aortic valve stenosis based on heart sounds

Abstract Background Aortic stenosis is still one of the major causes of sudden cardiac death in the elderly. Noninvasive screening for severe aortic valve stenosis (AS) may result in early cardiac diagnostic leading to an appropriate and timely medical intervention. Purpose The aims of this study were 1) to develop an artificial intelligence to detect severe AS based on heart sounds and 2) to build an application to screen patients using electronic stethoscope and smartphones, which will provide an efficient diagnostic workflow for screening as a complementary tool in daily clinical practice. Methods We enrolled 100 patients diagnosed with severe AS and 200 patients without severe AS (no echocardiographic sign of AS [n=100], mild AS [n=50], moderate AS [n=50]). The heart sounds were recorded in 4000 Hz waveform audio format at the following 3 sites of each patient; the 2nd intercostal right sternal border, the Erb's area and the apex. Each record was divided into multiple data of 4 seconds duration, which built 10800 sound records in total. We developed multiple convolutional neural networks (CNN) designed to recognize severe AS in heart sounds according to the recorded 3 sites. We adopted a stratified 4-fold cross-validation method by which the CNN was trained with 60% of the whole data, validated with 20% data and tested with the remaining 20% data not used during training and validation. As performance metrics we adopted the accuracy, F1 value and the area under the curve (AUC) calculated as the average of all cross-validation folds. For the smartphone application, we combined the best CNN-models from each recorded site for the best performance. Further 40 patients were newly enrolled for its clinical validation (no AS [n=10], mild AS [n=10], moderate AS [n=10], severe AS [n=10]). Results The accuracy, F1 value and AUC of each model were 88.9±5.7%, 0.888±0.006 and 0.953±0.008, respectively. The sensitivity and specificity were 87.9±2.2% and 89.9±2.4%. The recognition accuracy of moderate AS was significantly lower as compared to the other AS grades (moderate AS 74.1±6.1% vs no AS 98.0±1.4%, mild AS 97.6±1.2%, severe AS 87.9±2.2%, respectively, P<0.05). Our smartphone application showed a sensitivity of 100% (10/10), a specificity of 73.3% (22/30), and an accuracy of 80.0% (32/40), which implicated a good utility for screening. In the detailed analysis of 8 mistaken decisions, these were highly affected by the presence of severe mitral or tricuspid valve regurgitation despite of non-severe AS (7/8 [87.5%]). Conclusions This study demonstrated the promising possibility of an end-to-end screening for severe aortic valve stenosis using an electronic stethoscope and a smartphone application. This technology may improve the efficacy of daily medicine particularly where the human resource is limited or support a remote medical consultation. Further investigations are necessary to increase accuracy. Funding Acknowledgement Type of funding sources: None.

Download Full-text

On-device artificial intelligence: mobile solution for detecting severe aortic valve stenosis based on heart sounds

European Heart Journal ◽

10.1093/eurheartj/ehab724.3053 ◽

2021 ◽

Vol 42 (Supplement_1) ◽

Author(s):

H Makimoto ◽

T Shiraga ◽

B Kohlmann ◽

C.-E Magnisali ◽

R Schenk ◽

...

Keyword(s):

Artificial Intelligence ◽

Aortic Valve ◽

Aortic Valve Stenosis ◽

Cross Validation ◽

Performance Metrics ◽

Area Under The Curve ◽

Smartphone Application ◽

Medical Intervention ◽

Heart Sounds ◽

Electronic Stethoscope

Abstract Background Aortic stenosis is still one of the major causes of sudden cardiac death in the elderly. Noninvasive screening for severe aortic valve stenosis (AS) may result in early cardiac diagnostic leading to an appropriate and timely medical intervention. Purpose The aims of this study were 1) to develop an artificial intelligence to detect severe AS based on heart sounds and 2) to build an application to screen patients using electronic stethoscope and smartphones, which will provide an efficient diagnostic workflow for screening as a complementary tool in daily clinical practice. Methods We enrolled 100 patients diagnosed with severe AS and 200 patients without severe AS (no echocardiographic sign of AS [n=100], mild AS [n=50], moderate AS [n=50]). The heart sounds were recorded in 4000 Hz waveform audio format at the following 3 sites of each patient; the 2nd intercostal right sternal border, the Erb's area and the apex. Each record was divided into multiple data of 4 seconds duration, which built 10800 sound records in total. We developed multiple convolutional neural networks (CNN) designed to recognize severe AS in heart sounds according to the recorded 3 sites. We adopted a stratified 4-fold cross-validation method by which the CNN was trained with 60% of the whole data, validated with 20% data and tested with the remaining 20% data not used during training and validation. As performance metrics we adopted the accuracy, F1 value and the area under the curve (AUC) calculated as the average of all cross-validation folds. For the smartphone application, we combined the best CNN-models from each recorded site for the best performance. Further 40 patients were newly enrolled for its clinical validation (no AS [n=10], mild AS [n=10], moderate AS [n=10], severe AS [n=10]). Results The accuracy, F1 value and AUC of each model were 88.9±5.7%, 0.888±0.006 and 0.953±0.008, respectively. The sensitivity and specificity were 87.9±2.2% and 89.9±2.4%. The recognition accuracy of moderate AS was significantly lower as compared to the other AS grades (moderate AS 74.1±6.1% vs no AS 98.0±1.4%, mild AS 97.6±1.2%, severe AS 87.9±2.2%, respectively, P<0.05). Our smartphone application showed a sensitivity of 100% (10/10), a specificity of 73.3% (22/30), and an accuracy of 80.0% (32/40), which implicated a good utility for screening. In the detailed analysis of 8 mistaken decisions, these were highly affected by the presence of severe mitral or tricuspid valve regurgitation despite of non-severe AS (7/8 [87.5%]). Conclusions This study demonstrated the promising possibility of an end-to-end screening for severe aortic valve stenosis using an electronic stethoscope and a smartphone application. This technology may improve the efficacy of daily medicine particularly where the human resource is limited or support a remote medical consultation. Further investigations are necessary to increase accuracy. FUNDunding Acknowledgement Type of funding sources: None.

Download Full-text

Multiple biomarker panel to screen for severe aortic stenosis: results from the CASABLANCA study

Open Heart ◽

10.1136/openhrt-2018-000916 ◽

2018 ◽

Vol 5 (2) ◽

pp. e000916 ◽

Cited By ~ 1

Author(s):

Sammy Elmariah ◽

Cian McCarthy ◽

Nasrien Ibrahim ◽

Deborah Furman ◽

Renata Mukai ◽

...

Keyword(s):

Aortic Valve ◽

Predictive Value ◽

Cross Validation ◽

Severe Aortic Stenosis ◽

Area Under The Curve ◽

Diagnostic Model ◽

Aortic Valve Area ◽

Academic Medical Centre ◽

Diagnostic Score ◽

Biomarker Panel

ObjectiveSevere aortic valve stenosis (AS) develops via insidious processes and can be challenging to correctly diagnose. We sought to develop a circulating biomarker panel to identify patients with severe AS.MethodsWe enrolled study participants undergoing coronary or peripheral angiography for a variety of cardiovascular diseases at a single academic medical centre. A panel of 109 proteins were measured in blood obtained at the time of the procedure. Statistical learning methods were used to identify biomarkers and clinical parameters that associate with severe AS. A diagnostic model incorporating clinical and biomarker results was developed and evaluated using Monte Carlo cross-validation.ResultsOf 1244 subjects (age 66.4±11.5 years, 28.7% female), 80 (6.4%) had severe AS (defined as aortic valve area (AVA) <1.0 cm2). A final model included age, N-terminal pro-B-type natriuretic peptide, von Willebrand factor and fetuin-A. The model had good discrimination for severe AS (OR=5.9, 95% CI 3.5 to 10.1, p<0.001) with an area under the curve of 0.76 insample and 0.74 with cross-validation. A diagnostic score was generated. Higher prevalence of severe AS was noted in those with higher scores, such that 1.6% of those with a score of 1 had severe AS compared with 15.3% with a score of 5 (p<0.001), and score values were inversely correlated with AVA (r=−0.35; p<0.001). At optimal model cut-off, we found 76% sensitivity, 65% specificity, 13% positive predictive value and 98% negative predictive value.ConclusionsWe describe a novel, multiple biomarker approach for diagnostic evaluation of severe AS.Trial registration numberNCT00842868.

Download Full-text

Abstract 15206: Stethaid: Automated Point-of-care Recognition of Innocent Still’s Murmur in Children

Circulation ◽

10.1161/circ.142.suppl_3.15206 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Robin Doroshow ◽

Titus John ◽

Ravi Tej Ambati ◽

Raj Shekhar

Keyword(s):

Mobile Technology ◽

Learning Algorithm ◽

Point Of Care ◽

Area Under The Curve ◽

Smartphone Application ◽

The United States ◽

High Accuracy ◽

High Rate ◽

Electronic Stethoscope ◽

Heart Murmurs

Introduction: Still’s murmurs are difficult to distinguish from other heart murmurs at the primary care level. This leads to an estimated 400,000 children being referred to pediatric cardiologists each year for evaluation of heart murmurs in the United States. The murmur is ultimately diagnosed as innocent Still’s murmur in approximately 90% of these children. This diagnosis requires no specialty referral, cardiac testing, exercise restrictions or cardiology follow-up. These unnecessary referrals and associated tests waste healthcare resources, add to healthcare costs, and are a source of avoidable anxiety in children and families while waiting to see a pediatric cardiologist. Hypothesis: We have developed a novel mobile technology to discriminate Still’s murmur from all other murmurs. A smartphone application (app) that could quickly distinguish Still’s murmur from all other murmurs with high accuracy at the point of care could reassure pediatricians in their identification of Still’s murmur, significantly reducing the current rate of unnecessary referrals to cardiologists. Methods: Our current prototype is StethAid, a smartphone app which accepts heart sound recordings, and a cloud-based deep learning algorithm for discriminating Still’s murmur from other murmurs. The algorithm was independently developed and evaluated using archived heart sounds, recorded by one of the authors (RWD), made using an electronic stethoscope (3M Littmann Model 4100). Results: Our algorithm offers a sensitivity (Still’s murmurs correctly identified) of 90% with a specificity (Non-Still’s murmurs correctly identified) of 99% on 5-fold cross-validation over the Murmur Library. The area under the curve was 0.99446. The algorithm’s result is available in real time at the point-of-care (&lt 1 min). Conclusions: The described point-of-care mobile technology could automatically distinguish Still’s murmur from all other murmurs with high accuracy. The technology could lower the current high rate of specialist referrals associated with Still’s murmur and reduce the related financial and emotional costs. Future directions will include further improvement of the technology and validation through multi-center clinical trials.

Download Full-text

Diagnosis of aortic valve stenosis by correlation analysis of wavelet filtered heart sounds

Proceedings of the 25th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (IEEE Cat. No.03CH37439) ◽

10.1109/iembs.2003.1280518 ◽

2004 ◽

Cited By ~ 2

Author(s):

A. Voss ◽

J. Herold ◽

R. Schroeder ◽

F. Nasticzky ◽

A. Mix ◽

...

Keyword(s):

Aortic Valve ◽

Correlation Analysis ◽

Aortic Valve Stenosis ◽

Heart Sounds

Download Full-text

Hybrid feature selection approach to identify optimal features of profile metadata to detect social bots in Twitter

Social Network Analysis and Mining ◽

10.1007/s13278-021-00786-4 ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Eiman Alothali ◽

Kadhim Hayawi ◽

Hany Alashwal

Keyword(s):

Feature Selection ◽

Random Forest ◽

Cross Validation ◽

Performance Metrics ◽

Area Under The Curve ◽

Support Vector ◽

Detection Systems ◽

Vector Machines ◽

Selection Approach ◽

Feature Selection Approach

AbstractThe last few years have revealed that social bots in social networks have become more sophisticated in design as they adapt their features to avoid detection systems. The deceptive nature of bots to mimic human users is due to the advancement of artificial intelligence and chatbots, where these bots learn and adjust very quickly. Therefore, finding the optimal features needed to detect them is an area for further investigation. In this paper, we propose a hybrid feature selection (FS) method to evaluate profile metadata features to find these optimal features, which are evaluated using random forest, naïve Bayes, support vector machines, and neural networks. We found that the cross-validation attribute evaluation performance was the best when compared to other FS methods. Our results show that the random forest classifier with six optimal features achieved the best score of 94.3% for the area under the curve. The results maintained overall 89% accuracy, 83.8% precision, and 83.3% recall for the bot class. We found that using four features: favorites_count, verified, statuses_count, and average_tweets_per_day, achieves good performance metrics for bot detection (84.1% precision, 81.2% recall).

Download Full-text