Rhythm AI

World’s No.1 cardiac rhythm interpretation and reporting services company with deep domain expertise and proven capabilities. We drive greater profitability, operational excellence, improved patient care and business agility.


Model Validation

In order to understand and visualise the important regions in the ECG image used for prediction, a Grad-CAM analysis was performed on a dataset of 235 images.

The Grad-CAM algorithm, which works as a wrapper over a trained model, was applied on the dataset to generate a 299 x 299 image representing a heat map of class activation
mappings. The regions with high activations in the heat map were localized by extracting coordinates of pixels in the RGB colour space with the value in the R channel greater than 230. A bounding box was then drawn around the region with high activation mapping and another to map the coordinates to the original image for better visualization of the result.

Over the annotated external dataset, {x number} AFIB, {x number} PVCs, {x number} Ventricular Tachycardias and  {x number} concurrent arrhythmias {contained arrhythmias} were considered and the Grad-CAM algorithm was applied for visualization of important regions of the image


The overall model accuracy was 92% with a precision of about 91% for all classes except PSVT, junctional rhythm and advanced heart block, which showed an overall precision of 74.6%. The model classified 96% of all the tachycardias files, 90% of various types of AV blocks and 88% of the 152 artifact samples. Critical arrhythmias like atrial fibrillation, atrial flutter, ventricular tachycardia and PSVT have a recall of 0.81. 0.89. 0.95 and 0.85. More than 5 different devices were used to collect the test data and the performance model was satisfactory considering the signal quality and artifacts involved during the acquisition. The pre-processing steps, backed by source independent behaviour of the model, considerably improved the scalability in the practical scenario.

Be it direct digital data feed or paper records, DeeprrhythmiaTM makes it incredibly fast and easy to get results with high accuracy ensuring the right diagnosis at the right time

We automated arrhythmia detection with an aim to achieve accuracy greater than an average cardiologist by implementing an end-to-end solution with a robust pipeline to streamline the data from a multitude of sources to feed a deep learning model. The proposed image classification model detects 21 different arrhythmias and conduction abnormalities across a diversified set of medical devices well established in the industry today such as wearables (wristbands and smartwatches) and other continuous monitoring devices (holters and ECG patches). Signal to noise ratio of wearables that collect limb lead or similar signals from wrists or fingertips were also included in the model. Although deep learning is increasingly being adopted and used in the healthcare industry a lack of annotated data and validation strategies have limited the ability for conclusive interpretation of ECG detection models. Therefore, apart from external validation, a Gradient-weighted Class Activation Mapping was also used to generate visual explanations for model predictions so that the model no longer remains as a blackbox

Request For Services

Find out more about how we can help your organization navigate its next.
Let us know your areas of interest so that we can serve you better.

Click here

Further reading on more Products

Contact us

We're helping some of the most respected names in healthcare deliver measurably better outcomes. Let us show you what personally Human & AI integrated solution can do for your organization. While filling the form, please fill in the information more specifically that you are looking for.

Thank you for your query! We will get back to you shortly!!