June 17, 2025
Day 16- Deep Learning Model Development for ECG Analysis
What I Learned
Today at my internship, I made solid progress on our AI cardiovascular project. I continued refining the model training pipeline for ECG-based disease detection using three types of preprocessed data: time-domain features, frequency-domain components, and spectrograms. With all datasets already split into train, validation, and test sets, I began building and testing the deep learning models.Worked on developing the 1D CNN architecture for time-domain waveform classification. Planned the architecture for the 1D CNN + Transformer hybrid model using the frequency-domain dataset.Reviewed the input format and structure for 2D CNN training on spectrogram images. Also spent time understanding model performance metrics and preparing to log results for comparison.
Blockers
No Blockers
Reflection
Today was a day of laying important groundwork. With all the preprocessing behind me, I finally got to focus on building the actual deep learning models — the part I’ve been anticipating for a while. It felt good to shift from cleaning and prepping data to designing and testing model architectures.Starting with the 1D CNN for the time-domain ECG signals, I had to think carefully about the structure — balancing depth, filter sizes, and avoiding overfitting too early. At the same time, I began planning the 1D CNN + Transformer hybrid for the frequency-domain data. That part excited me the most, knowing it could potentially capture more temporal relationships in the signals.What stood out most today, though, was how everything started connecting. Preprocessing steps that felt tedious last week suddenly made sense now that I was using the clean datasets for actual training. I still have a lot to do — model tuning, evaluation, explainability. But today reminded me that real progress sometimes looks like quiet, focused effort behind the scenes. I’m building a system that could one day help detect heart conditions more accurately. That thought kept me motivated.