Picked from the software, topic, and level of “Class Classification of EKG Aritmia Wavelet Deep Learning #6: Class-wise Evaluation and Model Interpretation”.
At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Here is the final result, a comparison of four per-class arrhythmia models on a new patient, and a saliency map showing the parts of the heartbeat that the CNN pays attention to. The winner isn't the most complex one, and the model indeed pays attention to the complex QRS like a doctor. Let's get started. Welcome to the final episode of the EKG classification series with Wavelet and Deep Learning. In the medical field, models must be evaluated per-class and be explainable, because doctors won't trust a black box. Today we are comparing all models comprehensively and looking into the decisions made by the CNN.
Step 1: Class-wise Comparison
Create an evaluation underscore ek.py file. We load the predictions of the four models on DS2, then calculate accuracy, sensitivity and positive predictivity for each class, as well as F1 macro, the average F1 of the four classes which gives equal weight to the rare classes. Run. SVM achieves the highest F1 macro, 0.517, followed by random forest 0.492, while the two CNNs are around 0.46 to 0.48. All models detect around 90% of bit V, but bit S and F remain a major challenge in inter-patient evaluation. This is the honest picture that is often hidden behind the 99% accuracy in papers that mix patients. Continue with the interpretation.
Step 2: Saliency Map & Graph
We load the 1D CNN, then the saliency function calculates the gradient of the class score against each bit sample. Samples with large gradients most influence the decision. We plot the sensitivity diagram, positive predictivity, confusion matrix of the best model, F1 macro, and saliency 1 bit from each class. Let's take a look at the graphs. On the saliency map, dark red points pile up around the complex QRS near the R peak, and for bit V also in the part of the inverted T wave, In line with how cardiologists read EKG. The model learns medically sound information, but the time information from RR remains key for bit S.
Done!
Done, and the series is complete. We have built a complete research on EKG arrhythmia classification, MITBIH base data and inter-patient division, filtering and segmentation, wavelet and scalogram, SVM and random forest, CNN 1 and 2 dimensional, up to class-wise evaluation and sensitivity map. For further research, try adapting per-patient with a little labeled bit, bit sequence models like LSTM or transformer, EEG data for seizure detection, or implementation on wearable devices. Thank you for following this series. Currently working on a thesis, dissertation, or research in the field of biomedical engineering and artificial intelligence? Cerdas Inovasi Digital helps implement biomedical signal processing, from EKG signal filtering and segmentation, wavelet transformation, classic machine learning, up to CNN 1 and 2 dimensional along with its interpretation as well as opening a Python programming course for electrical and biomedical engineering. Want to master the entire series? From MITBIH base data to class-wise arrhythmia evaluation? All 6 episodes, complete with step-by-step PDF guides, dataset, and source code, are available for premium tutorial subscribers. Subscribe starting at Rp69.000 per month, save 28% with the 12-month package, or one-time payment of Rp999.000 for lifetime access. Want just one class? Buy the single class with lifetime access. Subscribe now at the premium page cerdasinovasidigital.com. Contact our WhatsApp, or visit cerdasinovasidigital.com. Thank you for watching, see you in the next episode. Thank you for watching,
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,290 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
