Apollo Hospitals, in collaboration with Google Health, brings about a groundbreaking
transformation in cardiovascular health, leveraging AI-driven DICOM ECGs to enhance diagnostic
accuracy and deliver personalized patient care globally.
Apollo Hospitals has pioneered a groundbreaking approach to leverage DICOM Electrocardiography
(ECG) ../images stored in institutional PACS servers. The objective is to establish a standardized
database using Standard Communication Protocol (SCP) and develop a precise Large Language Model
(LLM) capable of interpreting various cardiovascular diagnoses and cardiovascular disease
progression. Apollo Hospitals operates more than 70 hospitals serving more than 200 million
patients across India.
The study employs PyDicom libraries to retrieve, anonymize, and label DICOM ECGs, transforming
them into (x, y) coordinates. The resulting tabular format includes lead definitions, duration
and amplitude dimensions, metadata, clinical conditions, and textual diagnoses. An eXtreme
Gradient Boosting (XGBoost) model is trained on a pilot dataset of several thousand ECGs,
considering age, gender, clinical categories, and heart rate, demonstrating promising results
with a Cross Correlation Percentage of 0.92 and an AUC of 0.94 for predicting binary clinical
categories.
The AI-driven conversion of DICOM ECGs demonstrates substantial potential for improving
cardiovascular diagnoses and disease progression tracking. Apollo Hospitals envisions leveraging
these advancements to develop Large Language ECG Models, contributing significantly to enhanced
diagnostic accuracy and personalized patient care in cardiovascular health. Apollo’s AI-driven
diagnosis and risk stratification tools, extensively reviewed and validated before use, are now
being used in at least eight countries and have been adapted for other non-communicable
diseases, such as cardiovascular disease, diabetes, asthma, and liver fibrosis.
