This commit contains the submit from Cristina de la Torre for Idoven data scientist job offering.
The electrocardiogram (ECG) is a non-invasive representation of the electrical activity of the heart from electrodes placed on the surface of the torso. The standard 12-lead ECG has been widely used to diagnose a variety of cardiac abnormalities such as cardiac arrhythmias, and predicts cardiovascular morbidity and mortality. The early and correct diagnosis of cardiac abnormalities can increase the chances of successful treatments. However, manual interpretation of the electrocardiogram is time-consuming, and requires skilled personnel with a high degree of training.
Automatic detection and classification of cardiac abnormalities can assist physicians in the diagnosis of the growing number of ECGs recorded. Over the last decade, there have been increasing numbers of attempts to stimulate 12-lead ECG classification. Many of these algorithms seem to have the potential for accurate identification of cardiac abnormalities. However, most of these methods have only been tested or developed in single, small, or relatively homogeneous datasets.
- To know Physionet, and more specifically the PTB-XL Dataset.
- To know specific libraries related to Physionet to load, plot.. signals.
- To investigate more about ECG and the relevant paper that it has in the heart issues detection.
- To be more confident with the ECG area.
- To apply the knowledge acquired during my master studies in the subject Biosignals and Machine Learning.
During this task, I has been able to extract some features from ECG signal, that later will be relevant to detect some anomalies. Anomalies as for example Arrhythmia or any other heart disease.
We were able to see how the way we are working with the data can affect highly the time needed to train the models. In this case, working with spectograms instead of the raw ECG data can affect a lot the time needed.
Train and predict the spectogram in 0 sec (comvert all ECGs to spectograms 2sec). Train and predict the ECG in 40 sec. As we can see, this seems to not be a huge difference, but we have to take into account that we are working with a small part of the data and classe,s so if in future we will use the whole data this can affect a lot the time needed. Also it is really important to note that additional step to pass the raw ecg to the spectogram it took 2 sec for all the data we were working with.
In the other hand, if we compare the accuracy with the exact same model, we can see that it is better with the spectogram signal instead of with the raw ecg data. Another reason apart from the time, it is added to choose the spectogram over the raw ecg.
- Test more elaborated and individualized mother machine/deep learning models
- Work with all classes
- Use ECG images instead of singals to train models: This is another point I was thinking of. If doctor uses the ECG plot to detect illness, why not to follow the same path with deep learning?
- How to introduce Machine Learning/Deep Learning in future steps:
At first, introduce Machine Learning/Deep Learning to detect automatically the different peaks. In that way we will be able to later on use another Machine Learning/Deep Learning model to detect heart anomalies.
But, how? This last model as first propposal I will do the following:
Feature Selection: First of all, I will see the correlation between all the features of each signal (age, heart beat, R peak distance...) and the disease/anomalie. In that way, I will be able to see which features are the most relevant in the anomalie detection and we can delete those features that can be adding noise to our model.
Train data, test data and validation data. After evaluating the data that PTB-XL is providing us we can see that there is patients with different anomalies. So, with the aim of obtaining a more accurate results. It is really important to have a balanced data in the model, in other words to have equitative the quantity of patients with each disease in the different sets of data.
Once all these steps have been carried out, we can start to try the different machine learning/deep learning models that exhist nowadays and are in the benchmark. Thank you to those models the patient and doctors will have a positive impact in their lives. As Doctor, this will help to easing the workload giving them more time to focus on the patient. As patient, maybe they will have a faster result of their illness so It can be treated in a fastest way.
Pan, J. and Tompkins, W., 1985. A Real-Time QRS Detection Algorithm. IEEE Transactions on Biomedical Engineering, BME-32(3), pp.230-236
Wagner, P., Strodthoff, N., Bousseljot, R., Samek, W., & Schaeffter, T. (2022). PTB-XL, a large publicly available electrocardiography dataset (version 1.0.3). PhysioNet. Link
Subject Biosignals and Bioimages. Master in Information Health Engineering (2019-2020).
Github repositories: