I recently has the pleasure of developing a PQRST delineation algorithm for HeartEye. I got to revive my original PQRST Paper from 1997 and reimplement it for the 12 lead HeartEye signal together with Bram van Es.
This algorithm is geometric, and doesn’t use AI. However, AI tools like Cursor and Claude were excellent development partners for code generation, translation, and especially for insightful plots.
The results were satisfyingly accurate. P and T waves were detected quite well
Irregular rhythms like atrial fibrillation were clearly distinguished by RR and PQ variability.
The “banana” plots rank each dataset’s recordings by their variability score (ascending, normalised to 0–1). The separation is striking — every AFib recording sits far above the sinus-rhythm population, with very little overlap:
