1. 1/30 Classification of acute myocardial infarction based on discriminant analysis and automatic fiducial point detection in the ECG Group 856b Dept. of Health science and Technology Ina Lewinsky Mads Hylleberg Flemming H Gravesen Fall 2004

2. 2/30 Contents Introduction (Ina) Preprocessing (Ina) Feature selection and extraction (Mads) Fiducial point detection (Mads) Classification (Flemming) Discussion (Flemming)

3. 3/30 Introduction Initiating problem AMI Development Diagnosis Treatment STEMI vs. NSTEMI Previous studies Hypotheses Data Preprocessing

4. 4/30 Initiating problem 2000-3000 cases of AMI each year Fatal condition 10-20 % die before admission to hospital 20 % of these die during the stay Treatment gives better results when initiated earlier Fast and accurate diagnosis is needed to improve outcome Usefull in pre- (ambulance) Descision support especially for young learning doctors Definition of AMI Acute necrosis in myocardial structure based in luminal obstruction of coronary arteries Obstruction is mainly caused by athersclerosis and thrombosis

5. 5/30 Diagnosis of AMI World Health organisation Chest discomfort Rise in certain blood markers Creatine Kinase Myoglobin Troponin Typical ECG patterns Caused by ischemia or necrosis due to the obstruction

6. 6/30 Acute coronary syndrome

7. 7/30 Development of AMI Luminal obstruction is caused by atherosclerosis and thrombosis

8. 8/30 Two types of AMI ST elevation AMI (STEMI) Non ST elevation AMI (NSTEMI)

9. 9/30 Treatment Thrombolysis Dslfkh Sdkhf Only significant in ST elevation AMI Percutaneous intervention

10. 10/30 Previous studies Rule based systems Artificial neural networks Statistic approaches According to Willems et al. Statistical aprocahes yield best results Features Traditional features ST elevation 80 ms after J point T inversion Q wave Additional features Morphology of ST segment and T wave Reciprocal changes

11. 11/30 Hypotheses The use of reciprocal features and morphology features from the ST segment and T wave imporves the detection of AMI relative to the use of non traditional ST features alone It is possible in the classification to distinguish between the two groups : non ST elevation AMI and ST elevation AMI.

12. 12/30 Data Definition of groups: NSTEMI STEMI Healthy controls

13. 13/30 Preprocessing Noise in the ECG Power line noise Base line drift Electrode contact noise Purpose of filtering Attinuate noise to achieve a signal ready for fidoucial point detection Ensure the morphology of the ECG is intact

14. 14/30 Choice of filters Simple frequency selective filters are chosen Well proven apporach for ECG Easy to implement No problematic noise in the ECG Baseline filter High pass-implemented as low pass and subtracted from the signal IIR Cut off frequency of 0.67 Hz Low pass filter – remove high frequency noise FIR Cut off frequency of 40 Hz????? Notch filter – remove power line noise Bandstop FIR filter

15. 15/30 Result of preprocessing (averaging ???)

16. 16/30 Contents Introduction (Ina) Preprocessing (Ina) Feature selection and extraction (Mads) Fidoucial point detection (Mads) Classification (Flemming) Discussion (Flemming)

17. 17/30 Contents Introduction (Ina) Preprocessing (Ina) Feature selection and extraction (Mads) Fidoucial point detection (Mads) Classification (Flemming) Discussion (Flemming)

18. 18/30 JSTRQPP