Better patient outcomes using AI
The application database had a huge volume of past ruled out recordings. This data was analyzed to understand what kind of recordings were ruled out and why they were ruled out. Also, the application has a functionality using which an ECG technician can specify why a recording is being ruled out. This method was also used to generate datasets to analyze and identify the reason for ruling out. Such datasets were then used to build a Machine Learning model which could rule out recordings before technician’s triaging operation.
Different models were tried out based on Random Forest classification, Gradient Boosting algorithm, Deep Learning 1D Convolutional layers, LSTMs and Ensemble of Ensembles technique. Finally, Deep Learning 1D Convolutional architecture with different layers was selected based on Accuracy, Precision, Recall and F1 Score metrics.
The new enhanced solutions helped improve processes and increased accuracy in executing tasks. Applications were built to scale up and incorporate new enhancements and business changes in a rapid fashion.
With the Machine Learning model in place, we were able to bring in around 23% improvement in the volume of procedures reviewed by a technician in a day.6500 procedures per day improved to 8000 procedures.
