A Soft Computing Approach to Acute Coronary Syndrome Risk Evaluation

    

    

    Figure 1:  A fragment of the knowledge base for Acute Coronary Syndrome Risk Evaluation.
    
    

The issues of ECG and Related Clinical Manifestations tables, the columns Physical Exercise and Healthy Diet of Life Style Habits table and the columns Previous ACS, Chest Pain and Heredity of Patient’s Information table are filled with 0 (zero) or 1 (one) denoting, respectively, no or yes. In Cardiac Biomarkers table 0 (zero) and 1 (one) stand, respectively, for negative or positive test. The column Smoking Habits of the Life Style Habits table is populated with 0 (zero), 1 (one) or 2 (two), denoting Current Smoker, Ex-Smoker and Never Smoked, respectively. The values presented in the STE, Cardiac Biomarkers, LSH and RCM columns of ACS Predisposing table are the sum of the correspondent tables, ranging between [0, 2], [0, 3], [0, 4] and [0, 6] respectively. The domain of Body Mass Index (BMI) column is in the range [0, 2], wherein 0 (zero) denotes BMI < 25; 1 (one) stands for a BMI ranging in interval [25, 30]; and 2 (two) denotes a BMI = 30. The BMI was evaluated using the equation

BMI = Body Mass / Height² [30],

While in the Gender column 0 (zero) and 1 (one) stand, respectively, for female (F) and male (M).

Applying the algorithm presented in [31], to all the fields that make the knowledge base for ACS predisposing (Figure 1) and looking to the DoC_s values obtained in this manner, it is possible to set the arguments of the predicate referred to below, that also denotes the objective function with respect to the problem under analyze.

acs_predisposing : Age, Gen_der, Her_edity, Prev_{ious ACS}, C_hest P_ain, ECG, C_ardiac B_iomakers, B_ody M_ass I_ndex, L_ife S_tyle H_abits, R_elated C_linical M_{anifestations} → {0,1}

where 0 (zero) and 1 (one) denote, respectively, the truth values false and true.

Exemplifying the application of the algorithm presented in [31], to the patient that presents feature vector (77, 1, 1, 0, ⊥, 1, 1, 1, [0, 2], [2, 4]), one may get:

Begin,

The predicate’s extension that maps the Universe-of-Discourse for the term under observation is set ←

$\begin{array}{l}\{\\ \neg ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ \leftarrow ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ ac{s}_{diagnosis}\frac{(77,1,1,0,\perp ,1,1,1,\left[0,2\right],\left[2,4\right])}{\text{attribute`svalues}}\colon\colon 1\colon\colon DoC\\ \frac{\left[38,90\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,2\right]\left[0,3\right]\left[0,2\right]\left[0,4\right]\left[0,6\right]}{\text{attribute`sdomains}}\\ \}\colon\colon 1\end{array}$

The attribute’s values ranges are rewritten ←

$\begin{array}{l}\{\\ \neg ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ \leftarrow ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ ac{s}_{diagnosis}\frac{(\left[77,77\right],\left[1,1\right],\left[1,1\right],\left[0,0\right],\left[0,1\right],\left[1,1\right],\left[1,1\right],\left[1,1\right],\left[0,2\right],\left[2,4\right])}{\text{attribute`s values ranges}}\colon\colon 1\colon\colon DoC\\ \frac{\left[38,90\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,2\right]\left[0,3\right]\left[0,2\right]\left[0,4\right]\left[0,6\right]}{attribute`sdomains}\\ \}\colon\colon 1\end{array}$

The attribute’s boundaries are set to the interval [0, 1] ←

$\begin{array}{l}\{\\ \neg ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ \leftarrow ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ ac{s}_{predisposing}\frac{(\left[0.75,0.75\right],\left[1,1\right],\left[1,1\right],\left[0,0\right],\left[0,1\right],\left[0.5,0.5\right],\left[0.33,0.33\right],[0.5,0.5],\left[0,0.5\right],\left[0.33,0.67\right])}{\text{attribute`s values ranges once normalized}}\\ \colon\colon 1\colon\colon DoC\frac{\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]}{\text{attribute`s domains once normalized}}\\ \}\colon\colon 1\end{array}$

The DoC’s values are evaluated ←

$\begin{array}{l}\{\\ \neg ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ \leftarrow ac{s}_{predisposing}\left(Age,Gen,Her,Prev,CP,ECG,CB,BMI,LSH,RCM\right)\\ ac{s}_{predisposing}\frac{(1,1,1,1,0,1,1,1,0.87,0.94)}{\text{attribute`s confidence values}}\colon\colon 1\colon\colon 0.88\\ \frac{\left[0.75,0.75\right]\left[1,1\right]\left[1,1\right]\left[0,0\right]\left[0,1\right]\left[0.5,0.5\right]\left[0.33,0.33\right][0.5,0.5]\left[0,0.5\right]\left[0.33,0.67\right]}{\text{attribute`s values ranges once normalized}}\\ \frac{\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]\left[0,1\right]}{\text{attribute`s domains once normalized}}\\ \}\colon\colon 1\end{array}$

End.

The DoC is evaluated as it is illustrated in Figure 2, i.e., DoC = 1- Δl² . Here stands for the argument interval length, which was set in the interval [0, 1].

Figure 2: Evaluation of the Degree of Confidence.

    

    

    Figure 2:  Evaluation of the Degree of Confidence.
    
    

Close

Soft computing approach

The model to evaluate ACS predisposing risk presented previously demonstrates how all the information comes together and how it is processed. In this section, a soft computing approach to deal with the processed information is considered. Thus, a hybrid approach was set to model the universe of discourse, based on Artificial Neural Networks (ANNs), which are used to structure data and capture complex relationships between inputs and outputs [32-34]. ANNs simulate the structure of the human brain, being populated by multiple layers of neurons, with a valuable set of activation functions. As an example, let us consider the case given above, where one may have a situation in which the ACS diagnosis is needed. In Figure 3 it is shown how the normalized values of the attributes intervals boundaries and their DoC and QoI values work as inputs to the ANN. The output depicts the ACS predisposing risk, plus the confidence that one has on such a happening. In this study 1742 patients were considered. The ACS was diagnosed in 116 cases, i.e., in 6.7% of the analysed population Table 2. The coincidence matrix of the ANN model, where the values displayed denote the average of 30 experiments is present in Table 5. A glance of Table 5 shows that the model accuracy was 90.3% for the training set (1048 correctly classified in 1161) and 88.5% for test set (513 correctly classified in 581). Thus, the predictions made by the ANN model are satisfactory, attaining accuracies close to 90%. Based on coincidence matrix the values of sensitivity, specificity, PPV and NPV of the ANN model were computed for training and test sets Table 6. A perusal of Table 6 shows that the sensitivity ranges from 89.7% to 90.9%, while the specificity ranges from 88.4% to 90.2%.

Figure 3: The Artificial Neural Network Topology.

    

    

    Figure 3:  The Artificial Neural Network Topology.
    
    

Close

Table 5: The Coincidence Matrix for ANN Model.

  

    
Target
    
Predictive
  
  

    
Training set
    
Test set
  
  

    
True (1)
    
False (0)
    
True (1)
    
False (0)
  
  

    
True (1)
    
TP = 70
    
FN = 7
    
TP = 34
    
FN = 5
  
  

    
False (0)
    
FP = 106
    
TN = 978
    
FP = 63
    
TN = 479
  
  

    
Abbreviations:    FN - False Negative; FP - False Positive; TN - True Negative; TP - True    Positive 
  

Table 5:  The Coincidence Matrix for ANN Model.

Close

Table 6: Sensitivity, specificity, PPV and NPV for the ANN model.

  

    
 
    
Sensitivity (%)
    
Specificity (%)
    
PPV (%)
    
NPV (%)
  
  

    
Training set
    
90.9
    
90.2
    
39.8
    
99.3
  
  

    
Test set
    
89.7
    
88.4
    
35.7
    
99.2
  
  

    
Abbreviations: PPV - Positive Predictive Value; NPV - Negative Predictive Value 
  

Table 6:  Sensitivity, specificity, PPV and NPV for the ANN model.

Close

Discussion

Concerning the performance of cardiac biomarkers Table 4, the high sensitivity observed for cTnI (76.7%) and the high specificity observed for CK-MB (69.6%) can be related to the nature of such serum cardiac markers. Indeed, cTnI is an isoenzyme of Troponin I associated with fast-twitch cardiac muscle. Clinical studies have demonstrated that cTnI is detectable in the bloodstream 4-6 hours after an Acute Myocardial Infarction (AMI) and remains elevated for several days thereafter [35,36]. Further studies have indicated that cTnI has a higher clinical specificity for myocardial injury than does CK-MB [37]. Because of its cardiac specificity and sensitivity, cTnI has been used as a reliable marker in evaluating patients with unstable angina and NSTE-ACS [3,13]. Creatine kinase is a key enzyme of energy metabolism in muscle, catalyzing the reversible phosphorylation of creatine. The isoenzyme of creatine kinase found predominantly in cardiac muscle (i.e., CK-MB), presents low concentration in normal serum and non-cardiac tissues, whereby is the basis of its widely accepted use as a diagnostic indicator of myocardial injury [9]. Damage to the myocardium results in a transient release of CK-MB into the circulation, reaching a peak in 12 to 24 hours, crucial in the diagnostic of AMI. According to [9] about 98% of patients with AMI exhibit elevated serum CK-MB while only 3% of non-AMI patients exhibit abnormal CK-MB. Myoglobin shows lower specificity than the other markers. Myoglobin is the primary oxygen-carrying pigment of muscle tissue and is high when muscle tissue is damaged but it lacks specificity. This biomarker has the advantage of responding very rapidly, rising and falling earlier than CK-MB or troponin [10,11]. The performance of cardiac biomarkers based on the three age groups is similar to the observed for the whole population. Nevertheless, the sensitivity of biomarkers increases for the age group older than 65, which can be correlated with the high incidence of SCA observed for this age group. The specificity of all cardiac biomarkers is lower for this group as compared to the group between 35…50 years and also the 50… 65 one. The ROC curves for each cardiac biomarker are shown in Figure 4 and the AUC are presented in Table 7. The AUC, ranging between 0.556 and 0.732, are in concordance with the results of sensibility and specificity of biomarkers Table 4, since the cTnI exhibits the higher value of AUC and myoglobin presents the lower one. It denotes that the ability of the cardiac biomarkers to discriminate between diseased and nondiseased states is not very high, and should be complemented with other clinical information in order to conduce to a more effective diagnosis.

Figure 4: The ROC curves for cardiac biomarkers cTnI (?), CK-MB (− −) and Myoglobin (−··−) considering (a) All ages, (b) Age ∈ [35, 50], (c) Age∈[50, 65] and (d) Age > 65 years old.

    

    

    Figure 4:  The ROC curves for cardiac biomarkers cTnI (?), CK-MB (− −) and
Myoglobin (−··−) considering
(a) All ages, (b) Age ∈ [35, 50], (c) Age∈[50, 65] and (d) Age > 65 years old.
    
    

Close

Table 7: Areas under ROC Curves for the Cardiac Biomarkers.

  

    
 
    
All Ages
    
Age [35, 50[
    
Age[50, 65]
    
Age > 65 years old 
  
  

    
cTnI 
    
0.732
    
0.839
    
0.766
    
0.713
  
  

    
CK-MB 
    
0.649
    
0.693
    
0.631
    
0.650
  
  

    
Myoglobin
    
0.556
    
0.600
    
0.577
    
0.547
  
  

    
Abbreviations: CK-MB - Creatine Kinase MB;    cTnI - cardiac Troponin I 
  

Table 7:  Areas under ROC Curves for the Cardiac Biomarkers.

Close

Regarding the proposed ANN model, the sensitivity and specificity Table 6 are higher than the reported above for the cardiac biomarkers Table 4. The PPV ranges from 35.7% to 39.8% indicating that the proportion of cases labeled as diseased that were correctly classified obtained is not very high, and can be attributed to the low incidence of the disease in the cohort. Nevertheless, it is much higher than PPV values found for cardiac biomarkers. Finally, NPV ranges from 99.2% to 99.3% and are close to the values obtained for cardiac biomarkers. The ROC curves for the training and test sets are shown in Figure 5. The areas under ROC curves are higher than 0.9 for both cases, denoting that the model exhibits a good performance in recognition of ACS predisposing risk. The present model allows for the integration of the results of the cardiac biomarkers with other features such as heredity, previous ACS, ECG, life style habits and related clinical manifestations, allowing to be assertive in the diagnosis of ACS. This model showed a high sensibility, enabling the diagnosis of ACS comparing with the patients that really presented this pathology as well as classifying properly the absence of ACS (i.e., specificity). Thus it may be a major contribution to the early recognition and prevention of ACS.

Figure 5: The ROC curves for training set (?) and for test set (- -).

    

    

    Figure 5:  The ROC curves for training set (?) and for test set (- -).
    
    

Close

Conclusion

The diagnosis, decision support system presented in this study is able to give an adequate response to the need for a good method of ACS predisposing risk assessment. To go around the problem more effectively, other variables must be studied and considered, thus fulfilling important gaps in the existent risk assessment methods. Being an area filled with incomplete and unknown data it may be tackled by Artificial Intelligence based methodologies and techniques for problem solving. This work presents the founding of a computational framework that uses the powerful knowledge representation and reasoning techniques to set the structure of the information and the associate inference mechanisms. The knowledge representation and reasoning techniques presented above are very versatile and capable of covering almost every possible instance of the problem, namely by considering incomplete, contradictory, and even unknown data or knowledge, a marker that is not present in existing systems. Indeed, this method brings a new approach that can revolutionize prediction tools in all its variants, making it more complete than the existing methodologies and tools available. The model presented in this study showed a good performance in the detection of ACS predisposing, since their sensitivity and specificity exhibited values near 90%. These findings were corroborated by the area under ROC curves (> 0.9). This approach beyond allowing to obtain the ACS predisposing risk also lets us estimate the degree of confidence associated with this diagnosis (88.0% for the example presented above). Indeed, this is one of the added values of this approach, once it combines a framework based on Logic Programming to knowledge representation and reasoning, with a view to computing grounded on ANNs. Indeed, this approach encapsulates in itself a new vision to Multi-Value Logics, because once a proof of a theorem has been accomplished, it presents a measure of the confidence that the computational system has on such a proof.

References

1. Allender S, Peto V, Scarborough P, Kaur A, Rayner M. Coronary Heart Disease Statistics. 16th Edn. Oxford: British Heart Foundation. 2008.
2. Steg PG, James SK, Atar D, Badano LP, Blomstrom-Lundqvist C, Borger MA, et al. ESC guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation. Eur Heart J. 2012; 33: 2569-2619.
3. Roffi M, Patrono C, Collet J-P, Mueller C, Valgimigli M, Andreotti F, et al. 2015 ESC guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J. 2015; 37: 267-315.
4. Brogan RA, Malkin CJ, Batin PD, Simms AD, McLenachan JM, Gale CP. Risk stratification for ST segment elevation myocardial infarction in the era of primary percutaneous coronary intervention. World J Cardiol. 2014; 6: 865-873.
5. Fox KA, Eagle KA, Gore JM, Steg PG, Anderson FA; GRACE and GRACE2 Investigators. The Global Registry of Acute Coronary Events, 1999 to 2009--GRACE. Heart. 2010; 96: 1095-1101.
6. Terkelsen CJ, Lassen JF, Norgaard BL, Gerdes JC, Jensen T, Gotzsche LB, et al. Mortality rates in patients with ST-elevation vs. non-ST-elevation acute myocardial infarction: observations from an unselected cohort. Eur Heart J. 2005; 26: 18-26.
7. Anderson JL, Adams CD, Antman EM, Bridges CR, Califf RM, Casey-Jr, DE, et al. ACC/AHA 2007 Guidelines for the Management of Patients With Unstable Angina/Non-ST-Elevation Myocardial Infarction: Executive Summary. Circulation. 2007; 116: 803-877.
8. Cooper R, Cutler J, Desvigne-Nickens P, Fortmann SP, Friedman L, Havlik R, et al. Trends and disparities in coronary heart disease, stroke, and other cardiovascular diseases in the United States: findings of the national conference on cardiovascular disease prevention. Circulation. 2000; 102: 3137-3147.
9. Antman E, Bassand J-P, Klein W, Ohman M, Sendon JLL, Rydén L, et al. Myocardial infarction redefined - A consensus document of The Joint European Society of Cardiology/American College of Cardiology Committee for the redefinition of myocardial infarction. Eur Heart J. 2000; 21: 1502-1513.
10. Hollander JE. The Future of Cardiac Biomarkers, new concepts and emerging technologies for emergency physicians. EMCREG International. 2005; 4: 1-7.
11. Mueller C. Biomarkers and acute coronary syndromes: an update. Eur Heart J. 2014; 35: 552-556.
12. Thygesen K, Mair J, Katus H, Plebani M, Venge P, Collinson P, Lindahl B. Recommendations for the use of cardiac troponin measurement in acute cardiac care. Eur Heart J. 2010; 31: 2197-2204.
13. Agewall S, Giannitsis E, Jernberg T, Katus H. Troponin elevation in coronary vs. non-coronary disease. Eur Heart J. 2011; 32: 404-411.
14. Neves J. A logic interpreter to handle time and negation in logic databases. Muller R, Pottmyer J, editors. In: Proceedings of the 1984 Annual Conference of the ACM on the 5^th Generation Challenge. Association for Computing Machinery. 1984; 50-54.
15. Neves J, Machado J, Analide C, Abelha A, Brito L. The halt condition in genetic programming. Neves J, Santos MF, Machado J, editors. In: Progress in Artificial Intelligence, Lecture Notes in Artificial Intelligence 4874. Springer. 2007; 160-169.
16. Cortez P, Rocha M, Neves J. Evolving Time Series Forecasting ARMA Models. J Heuristic. 2004; 10: 415-429.
17. Haykin S. Neural Networks and Learning Machines. 3^rd edition. New York: Prentice Hall. 2008.
18. Machado J, Abelha A, Novais P, Neves J. Quality of service in healthcare units. Bertelle C, Ayesh A, editors. In: Proceedings of the ESM 2008. Eurosis - ETI Publication. 2008; 291-298.
19. Cardoso L, Marins F, Magalh&aTilde;es R, Marins N, Oliveira T, Vicente H, et al. Abstract Computation in Schizophrenia Detection through Artificial Neural Network based Systems. The Scientific World Journal. 2015.
20. Lucas P. Quality checking of medical guidelines through logical abduction. Coenen F, Preece A, Mackintosh A, editors. In: Proceedings of AI-2003 (Research and Developments in Intelligent Systems XX). Springer. 2004; 309-321.
21. CLSI (formerly NCCLS). Procedures for the Handling and Processing of Blood Specimens for Common Laboratory tests - Approved Guideline. 4^th edition. Wayne: Clinical and Laboratory Standards Institute. 2010.
22. Hong T, Hart K, Soh L-K, Samal A. Using spatial data support for reducing uncertainty in geospatial applications. GeoInformatica. 2014; 18: 63-92.
23. Li R, Bhanu B, Ravishankar C, Kurth M, Ni J. Uncertain spatial data handling: modeling, indexing and query. Computers and Geosciences. 2007; 33:42-61.
24. Schneider M. Uncertainty management for spatial data in databases: Fuzzy spatial data types. Güting RH, Papadias D, Lochovsky F, editors. In: Advances in Spatial Databases, Lecture Notes in Computer Science, 1651. Springer. 1999; 330-351.
25. Freire L, Roche A, Mangin JF. What is the best similarity measure for motion correction in fMRI time series? IEEE Trans Med Imaging. 2002; 21: 470-484.
26. Kakas A, Kowalski R, Toni F. The role of abduction in logic programming. Gabbay D, Hogger C, Robinson I, editors. In: Handbook of Logic in Artificial Intelligence and Logic Programming, 5. Oxford University Press. 1998; 235-324.
27. Pereira LM, Anh HT. Evolution prospection. Nakamatsu K, Phillips-Wren G, Jain LC, Howlett RJ, editors. In: New Advances in Intelligent Decision Technologies: Results of the First KES International Symposium IDT, Studies in Computational Intelligence, 199. Springer. 2009; 51-63.
28. Florkowski C. Sensitivity, Specificity, Receiver-Operating Characteristic (ROC) Curves and Likelihood Ratios: Communicating the Performance of Diagnostic Tests. Clin Biochem Rev. 2008; 29: 83-87.
29. Hajian-Tilaki K. Receiver Operating Characteristic (ROC) Curve Analysis for Medical Diagnostic Test Evaluation. Caspian J Intern Med. 2013; 4: 627-635.
30. World Health Organization: Obesity and overweight.
31. Fernandes F, Vicente H, Abelha A, Machado J, Novais P, Neves J. Artificial Neural Networks in Diabetes Control. In: Proceedings of the 2015 Science and Information Conference. IEEE. 2015; 362-370.
32. Vicente H, Dias S, Fernandes A, Abelha A, Machado J, Neves J. Prediction of the Quality of Public Water Supply using Artificial Neural Networks. J Water Supply Res T. 2012; 61: 446-459.
33. Vicente H, Couto C, Machado J, Abelha A, Neves J. Prediction of Water Quality Parameters in a Reservoir using Artificial Neural Networks. Int J Des Nat Ecodyn. 2012; 7: 309-318.
34. Vicente H, Roseiro J, Arteiro J, Neves J, Caldeira AT. Prediction of bioactive compound activity against wood contaminant fungi using artificial neural networks. Can J Forest Res. 2013; 43: 985-992.
35. Mair J, Morandell D, Genser N, Lechleitner P, Dienstl F, Puschendorf B. Equivalent Early Sensitivities of Myoglobin, CK-MB Mass, CK Isoform Ratios and Cardiac Troponins I and T for AMI. Clin Chem. 1995; 41: 1266-1272.
36. Larue C, Calzolari C, Bertinchant JP, Leclercq F, Grolleau R, Pau B. Cardiac-Specific Immunoenzymometric Assay of TnI in the Early Phase of AMI. Clin Chem. 1993; 39: 972-979.
37. Keller T1, Zeller T, Peetz D, Tzikas S, Roth A, Czyz E, Bickel C. Sensitive troponin I assay in early diagnosis of acute myocardial infarction. N Engl J Med. 2009; 361: 868-877.

Download PDF

Citation: Vicente H, Martins MR, Mendes T, Vilhena J, Gra&nTilde;eda JM, Gusm&aTilde;o R, et al. A Soft Computing Approach to Acute Coronary Syndrome Risk Evaluation. Austin J Clin Cardiolog. 2016; 3(1): 1044. ISSN : 2381-9111

Home

Journal Scope

Online First

Current Issue

Editorial Board

Instruction for Authors

Submit Your Article

Contact Us