Predicting Troponin Elevation from Body Surface Cardiac Map ECG Variables and from Standard 12-Lead ECG
The majority of patients who suffer a “heart attack” do so in their own home and typically the alarm is raised by chest pain complain by the patent. The standard 12-lead ECG is a primary first detection method for a heart attack. However, it cannot “see” the whole surface of the heart and so around 50% of potential heart attacks have an inconclusive 12-lead ECG. Inconclusive cases remain undiagnosed until blood is analysed and the change of a particular blood enzyme such as troponin can be detected, taking up to 12-hours; time during which the heart attack is remaining untreated and heart muscle is being damaged. We cannot treat all chest pains as if they are heart attacks (i.e. before a confirmed diagnosis) as there are many other disease conditions that give rise to heart attack-like chest pain, and the treatments for a heart attack have their own risks. Our multidisciplinary laboratory has conducted investigations since 1995 on the 80-lead body surface cardiac map (BSCM-80). This has resulted in improved detection of acute myocardial infarction (AMI) by approximately 20% whilst maintaining a high specificity of over 90% mainly by increasing detection of ischemia in the high right anterior, posterior, right ventricular and left bundle branch regions. This is not achievable with the gold standard 12-lead ECG, particularly within the first hour of the onset of symptoms. In order to make the BSCM-80 more portable, user-friendly and therefore with improved accessibility within the pre-hospital and Accident and Emergency departments, key information on both the anterior and posterior torso surfaces has been reduced into our newly proposed 28-lead BSCM system (BSCM-28) [1]. All BSCM-80 data in our trials to date have been generated in humans. We now have a very large database of chest pain patients with > 5,000 patients. Also, a database of maps has been collected in the pre-hospital coronary care unit operating from Royal Victoria Hospital Belfast using a portable prototype 28-lead system. The available BSCM-28 database has 400 patients. In data mining methods, artificial neural networks are particularly well-suited for quantitative structure-property and structure-activity relationships because of their ability to extract both linear and nonlinear information present in the mapping of physic-chemical descriptors to biological activity. Particularly, fuzzy neural network architectures, such as ARTMAP architectures [2,3], will be investigated for predicting troponin blood enzyme elevation from electrocardiographic variables extracted from the available BSCM-80 and BSCM-28 clinical databases and the accuracy of the method assessed against standard 12-Leas ECG. References 1) "Deriving a reduced lead system from the 80-lead body surface map in the electrocardiographic determination of acute myocardial infarction". Scott PJ, Stevenson M, Giardina M, Hamiltion A, Bennett JJ, Owens C, Manoharan G, Escalona O, Anderson J, Adgey AAJ. Journal of Electrocardiology; 41: 640-641, 2008. 2) “A new fuzzy ARTMAP approach for prediction of biological activity of potential HIV-1 protease inhibitors” Andonie R; Magill, L; Fabry-Asztalos, L, et al.. Proceedings of the IEEE International Conference on Bioinformatics and Biomedicine, Pages: 56-61, 2007. 3) “A Novel Fuzzy ARTMAP Architecture with Adaptive Feature Weights based on Onicescu's Informational Energy”, Andonie, R; Sasu, LM; Cataron, A. International Journal of Computers Communications & Control; Vol. 4: 104-117, 2009. 4) “Utilising a Genetic Algorithm to Minimise the Number of Leads in Body Surface Mapping for the Electrocardiographic Diagnosis of Myocardial Infarction”. Scott PJ, Navarro CO, Giardina M, Escalona OJ, Anderson JMcC, Adgey AAJ; Computing in Cardiology; 37:297−300, 2010. 5) “Optimization of the precordial leads of the 12-lead electrocardiogram may improve detection of ST-segment elevation myocardial infarction”. Scott PJ, Navarro C, Stevenson M, Murphy JC, Bennett JR, Owens C, Hamilton A, Manoharan G, Adgey AAJ; Journal of Electrocardiology; 44 (4): 425-431, 2011.
Personnel Involved
First Supervisor: Escalona, OJ Prof
Second Supervisor: McLaughlin, JAD Prof
Collaborator: Professor Razvan Andonie, Prof Jennifer Adgey (BHSCT)
Collaboration: The Royal Victoria Hospital Belfast, or The Ulster Hospital.
Synopsis:
The project will investigate novel artificial neural networks and associated rule extraction techniques for predicting troponin elevation from electrocardiographic variables. A comparative study of BSCM-80 and BSCM-28 will be carried out with respect to standard 12-Lead ECG methods, using available BSCM databases at our Research Institute.