Detection of myocardial ischemia due to clinically asymptomatic coronary artery stenosis at rest using supervised artificial intelligence-enabled vectorcardiography – A five-fold cross validation of accuracy




Coronary artery disease (CAD) is a leading cause of death and disability. Conventional non-invasive diagnostic modalities for the detection of stable CAD at rest are subject to significant limitations: low sensitivity, and personal expertise. We aimed to develop a reliable and time-cost efficient screening tool for the detection of coronary ischemia using machine learning.


We developed a supervised artificial intelligence algorithm combined with a five lead vectorcardiography (VCG) approach (i.e. Cardisiography, CSG) for the diagnosis of CAD. Using vectorcardiography, the excitation process of the heart can be described as a three-dimensional signal. A diagnosis can be received, by first, calculating specific physical parameters from the signal, and subsequently, analyzing them with a machine learning algorithm containing neuronal networks. In this multi-center analysis, the primary evaluated outcome was the accuracy of the CSG Diagnosis System, validated by a five-fold nested cross-validation in comparison to angiographic findings as the gold standard. Individuals with 1, 2, or 3- vessel disease were defined as being affected.


Of the 595 patients, 62·0% (n = 369) had 1, 2 or 3- vessel disease identified by coronary angiography. CSG identified a CAD at rest with a sensitivity of 90·2 ± 4·2% for female patients (male: 97·2 ± 3·1%), specificity of 74·4 ± 9·8% (male: 76·1 ± 8·5%), and overall accuracy of 82·5 ± 6·4% (male: 90·7 ± 3·3%).


These findings demonstrate that supervised artificial intelligence-enabled vectorcardiography can overcome limitations of conventional non-invasive diagnostic modalities for the detection of coronary ischemia at rest and is capable as a highly valid screening tool.


Coronary artery disease
Machine learning
Neural networks
Coronary ischemia

Short tweet summarizing the “take-home” message of the provided manuscript: Cardisiography – a non-invasive, highly valid screening tool for the detection of coronary ischemia at rest – combines a five lead vectorcardiography approach with supervised artificial intelligence. #cardisiography #ischemia.

View full text

Click Here To download full peer review ($37.50 fee required)

Cardisiography: a New Non-Invasive Procedure for Diagnosis of Hemodynamically Relevant Stenoses (≥ 50%) of the Coronary Arteries at Rest 

 Tenerich G, et al. 

 Mapping the spatial heterogeneity of coronary blood flow enables the detection of clinically unapparent resting ischemia. The regional differences of flow in individual areas of the myocardium, called micro-heterogeneities, were first described by Yipinsoi et al. (1) and confirmed by numerous investigations using different methods and in different species (2). A review of local myocardial perfusion using microspheres at a resolution of 300 µm shows that about one tenth of areas receive less than 50% of average flow (called low-flow areas), and a further one tenth receives more than 150% (high-flow areas). Blood flow within a particular area can vary by a factor of 10 (3). 

 Loncar et al. showed that high-flow areas do not represent luxury perfusion, but that local flow actually reflect the surrounding tissue‘s perfusion requirement (4). According to Austin et al. the resting blood flow is determined by metabolic factors, while the maximum flow is dependent on factors such as coronary pressure, diastole length and extra-luminal tissue pressure. For regions with a lower coronary reserve (such as coronary vascular disease), they postulate an increased vulnerability to ischemia, i.e. hemodynamically relevant stenoses defined as a narrowing of the lumen ≥ 50% (5). Regional differences in the duration of heart action potential are associated with the spatial heterogeneity of myocardial blood flow and can be determined using Activation Recovery Intervals (ARIs) via unipolar electrograms utilizing a neutral reference electrode (6) 

 Compared with the gold standard of coronary diagnostics, the coronary angiography whose specificity and sensitivity is 100%, non-invasive procedures such as the resting ECG and resting echocardiography show a significantly lower sensitivity and specificity (between 1-25%). Only during a stress test do these increase, for both methods, but this requires the attendance of a physician. Furthermore, both procedures are time-consuming (15-30 min), and the results of echocardiography are examiner-dependent. Other non-invasive procedures such as Myocardial Perfusion Syndrome (MPS), Coronary Computer Tomography (CCT) or Cardiac Magnetic Resonance Imaging (MRI) require considerable equipment and personnel expenditure and the presence of a physician, which also represents a significant cost (7,8,9,10,11,12). 


Cardiography is a non-invasive, examiner-independent, reproducible, fast, feasible, and cost-effective diagnostic procedure for hemodynamic detection of relevant coronary stenoses (≥50%) at rest. The diagnosis is made via a computer-based infinitesimal, three-dimensional calculation of the excitation processes of the mammalian heart based on a specific algorithm in conjunction with a neural network correlated to the intrinsic blood, as well as the specific spatial orientation of the myocardium in the dipole field as a function of time starting from a defined point Ɛ (13). 

For the purpose of this study, 182 patients’ records were correlated prospectively with the data of their respective coronary angiography: 89 men aged 39 to 84, and 93 female subjects aged 28 to 88. Individuals with 1/2/3- vascular disease with a stenosis grade of ≥ 50% were defined as being ill. 

Individuals without coronary artery disease or with stenosis <50% were defined as healthy. The Cardisiograph analysis was performed via the derivation of the excitation in the heart after placement of 4 electrodes (13) and connection to the Cardisiograph for a derivation period of 4 min. The results were correlated with corresponding coronary angiography test results, and for each group, sensitivity and specificity were calculated at a 95% confidence interval 


The sensitivity of 97.3% calculated for the male group (P (Se> 94.1%) = 95%), assuming a prevalence of 10%, corresponds to a negative predictive value (NPV) of 99%, meaning that disease can be ruled out with a high probability. The specificity of 62.5% (P (Sp> 42.6%) = 95%) corresponds to a Positive Predictive Value (PPV) of 22.4%, indicating that a positive test result requires further cardiological workup.

Figure 1) Left: Results for the entire male group, including training, validation, and test data; Right: Results exclusively for test data

For the female group, a sensitivity of 86% (Se> 75%) = 95%) at a false negative rate of about 14%, and a specificity of 75% (P (Sp> 62%) = 95%) at a false positive rate of about 25%, were calculated. In the case of a negative result, disease can be ruled out with a high probability due to the NPV of 97% (the PPV is 26.5%).

Figure 2) Left: Results for the entire female group, including training, validation and test data; Right: Results exclusively for test data


Existing non-invasive methods for detecting resting ischemia are either correlated with a very low sensitivity and specificity, or are costly, equipment- intensive efforts that require the presence of a physician. 

Cardisiography provides an alternative method: a non-invasive, cost-effective, reproducible, fast-to-perform and investigator-independent diagnostic procedure to detect hemodynamically relevant coronary vascular disease (stenosis ≥50%) at rest. 

For the male group, based on a sensitivity of 97.3%, Cardisiography can detect or rule out a coronary vascular disease with a high probability. In the female group, sensitivity is slightly lower at 86%, but its diagnostic value still compares favourably with existing complex, non-invasive procedures. For the male group, the diagnostic value of Cardisiography competes with the authority of coronary angiography in ruling out CVD. 


  1. Yipintsoi T, Dobbs WA, Jr., Scanlon PD Knopp TJ, Bassingthwaighte JB. Regional distribution of diffusible tracers and carbonized microspheres in the left ventricle of isolated dog hearts. Circ Res 1973; 33(5): 573-587 
  2.  Deussen A. Blood flow heterogeneity in the heart. Basic Res Cardiol 1998; 93(6):430-438 
  3.  Laussmann T, Janosi RA, Fingas CD, Schlieper GR, Schlack W, Schrader J et al. Myocardial proteome analysis reveals reduced NOS inhibition and enhanced glycolytic capacity in areas of low local blood flow. FASEB J 202; 16(6): 628 630 
  4.  Loncar R, Flesche CW, Deussen A. Coronary reserve of high-and low-flos regions in the dog heart left ventricle. Circ 1998; 98(3):262-270 
  5.  (5)Austin RE, Aldea GS, Coggins DL, Flynn AE, Hoffman JIE. Profound Spatial Heterogeneity of Coronary Reserve-Discordance between Patterns of Resting and Maximal Myocardial Blood-Flow. Circ Res 1990;67(2):319-331 
  6.  Millar CK, Kralios FA, Lux RL. Correlation Between Refractory Periods and Activation-Recovery Intervals from Electrograms – Effects of Rate and Adrenergic Interventions. Circ 1985: 72(6): 1372-1379 
  7.  Maltagliati A, Berti M, Muratori M, Tamborini G, Zavalloni D, Berna G, Pepi M. Exercise echocardiography versus exercise electrocardiography in the diagnosis of coronary artery disease in hypertension. Am J Hypertens 2000; 13:796-801 
  8.  Gargiulo P, Petretta M, Bruzzese D, Cuocolo A, Prastaro <m, D´Amore C, Vassallo E, Savarese G, Marciano C, Paolillo S, Filardi PP. Myocardial perfusion scintigraphy and echocardiography for detecting coronary artery disease in hypertensive patients: a meta-analysis. Eur J Nucl Med Mol Imaging 2011; 38:2040-2049 
  9.  Hadamitzky M, Meyer T, Hein F, Bischoff B, Byrne RA Martinoff S, Schömig A, Hausleiter J. Prognostic value of coronary computed tomographic antiography in patients with arterial hypertension. Int J Cardiovasc Imaging 2012; 28:641-650 
  10.  Koenig W, Khuseyinova N. Biomarkers of atherosclerotic plaque instability and rupture. Arterioscler Thromb Vasc Biol 2007; 27:15-26 
  11.  Astarita  C,  Pálinkás  A,  Nicolai  E,  Maresca  FS,  Varga  A,  Picano  E. Dipyridamoleatropine stress echocardiography versus exercise SPECT scintigraphy for detection of coronary artery disease in hypertensives with positive exercise test. J Hypertens 2001; 19:495- 502 
  12.  Milosavljevic  J,  Ostojic  M,  Marinkovic  J.  Dipyridamole-dobutamine stress echocardiography for the detection of myocardial ischemia in patients with hypertension. Herz 2005; 30:215-222 
  13.  Patent application: Cardisio 

Please click here to download a copy of the Abstract Cardisiography