Hi, I am trying to develop a signal classification model for ECG signals, using the PTB-XL dataset (PTB-XL, a large publicly available electrocardiography dataset v1.0.3). The dataset has a column called scp_codes, which indicates whether the signal is NORM (normal) or abnormal with the specific illness class. Table below shows list of abnormal classes and their distribution
The numbers of individual classes.
| Number of Records | Class | Description |
|---|---|---|
| 7185 | NORM | Normal ECG |
| 3232 | CD | Myocardial Infarction |
| 3064 | STTC | ST/T Change |
| 2936 | MI | Conduction Disturbance |
| 815 | HYP | Hypertrophy |
ECG is stored in 10 sec clips of signal reading, which is in a wfdb waveform format with a sample rate of 100 Hz. Example plot of data which I read from PTB-XL data looks like as shown below. It is already labeled as NORM or class of abnormality as defined by scp_code attached to the data.
I am trying to use the labeled signals in this dataset to classify future ECG signals. I have already built a CNN model, but I am hoping to use HuggingFace to increase my accuracy. So far, I am thinking of using the Audio Spectrogram Transformer for audio classification by converting the signal into a spectrogram. Do you think this method could work? Do you know what other models I could use for signal classification?