Benchmarks

Development and testing of the ECG Recognition Library was performed on CPSC 2018.

Differential diagnosis of MI and BER in case of significant ST-elevation

Evaluation metrics: balanced accuracy.

option	Accuracy
original	73%
tuned	82%

Classification between Normal and Abnormal classes

Performed with ecg_is_normal.

The model supports specifying the abnormality class for which subset data will be selected:

• STTC (ST/T Change)
• MI (Myocardial Infarction)
• HYP (Hypertrophy)
• CD (Conduction Disturbance)
• ALL (All available classes)

Accuracy and F1-score are used to evaluate each scenario.

Abnormality class	Accuracy	F1-score
STTC	89.75%	92.44%
MI	88.74%	91.55%
HYP	87.39%	91.6%
CD	91.69%	94.04%
ALL	89.04%	84.44%