Metti Paak

Respiration is an important biological marker for determining the state of health or for predicting an aberration or disease. It is not easy to measure respiration outside clinical settings as it requires inconvenient equipment and sensor attachments. Therefore, it is important to accurately estimate the respiration information from available data such as the ECG and PPG data. I conducted this project in two phases: (i) developing and testing against available data with ground truth and (ii) improving and modifying the program to act on unlabeled data. I completed the project using a variety of tools (data ingestion, feature engineering, deep/transfer learning, ensemble learning) which achieved the desired business target. In phase 2, I modified the model to use propriety data, I implemented a performance monitoring tool to assess potential model drift. This model can be used interactively and in a production environment (deployed on AWS).
In this two-phase project, I developed an object-oriented program to predict RR from ECG and PPG (single or multi-channel). The program ingests data from various sources and in different formats after data preprocessing (anomaly detection, outlier removal, imputation, and quality assessment).

For a startup, I developed an AI-based MVP hosted on GCP.

Use case: a retina image is uploaded, image segmentation identifies retina blood vessels, optic cup, and disk (UNet based). This is a multimodal system that also takes into account the eye inter-ocular pressure IOP. The model outputs are (i) a risk stratification score, (ii) patient monitoring longitudinal analysis, and (iii) eye-level information to ophthalmologists as a decision support tool.

I have performed a product-market fit (PMF) analysis, customer development, and user interviews to assess the feasibility and market size for a novel AI-based solution for disease prediction. I provided a pitch deck and presented the idea to venture partners and angel investors. The solution was designed to serve as a platform with an easy to use hardware component to evaluate patients at home or the point of care. I prepared an FDA road map, competitive landscape, and established partnerships.