AI-based calculating vital cardiovascular metrics

Business Goals

Healthcare and IOT startup focused on cardiac health wanted to develop a prototype system for calculation of vital cardiovascular metrics to test the business hypothesis.

Challenge

• Unstable data coming from the sensors. Missing data points, such as missing peaks and fiducial points, required proper data preprocessing.
• High level of noise present in the data from the sensors. Different modalities introduce different noise profiles, requiring the use of a variety of denoising strategies and tuning.
• Different natures for the fiducial points of different signal modalities had to be considered during the detection.
• A combination of signal modalities had to be examined at the same time to calculate PEP/LVET/IVCT metrics.

Results

• Developed an accurate PEP/LVET/IVCT estimation system for patients, validating the business hypothesis with optimal strategies used.
• Developed tailored a solution that works seamlessly with the data from the client’s hardware (sensors).

Implementation Details

• An R&D was conducted to quickly estimate the traditional, analytical approaches as well as AI-based approaches for both the data denoising and fiducial points detection.
• Traditional approaches offered acceptable results fast to validate business hypothesis in weeks notice. We developed a prototype that dealt with missing data, provided denoising for all the signal modalities, detected fiducial points and peaks, and calculated proposed metrics.
• The use of Artificial Intelligence-based solutions such as autoencoders for denoising and CNNs as well as LSTMs for fiducial points detection was proposed for the later stages of development.