Smartphone phenotype collection for diagnostic screening of mild cognitive impairment

Parabon Nanolabs, Inc., Reston VA

Investigators

Abstract

Project Summary This project addresses a critical need for early detection of mild cognitive impairment (MCI) and other Alzheimer's-related dementias (ADRD). Advances in smartphone hardware, computer vision, and machine learning have enabled the possibility of producing smartphone-based cognitive testing applications able to collect electronic sensor data and transform it into highly informative phenotypes that can serve as early indicators of future disease progression. In this project, we aim to develop a revolutionary new smartphone- based cognitive testing platform, called CTX, that will enable the rapid development and deployment of smartphone-based tests that can capture raw sensor streams in a synchronized fashion, subsample and compress the combined streams, and transmit them to a cloud server for subsequent analysis and modeling. CTX will provide a high-level application development framework that will significantly reduce the time and technical knowledge required to produce a smartphone-based cognitive testing application by providing an application programming interface (API) that enables developers to simply declare what sensor data should be collected and when. The framework will handle all the details of collecting the sensor data, synchronizing it, and transmitting it to a back-end server. The API will also have a variety of other high-level features to facilitate development of cognitive test apps. To demonstrate the feasibility of our vision for CTX, in Aim 1 of this project we will develop the software framework, back-end server software and a prototype smartphone app to exercise and validate many of the platform's features. For Aim 2, we will develop three different tests for this app to test saccade (eye movement) latency, verbal recall, and wrist mobility, each collecting a different type of sensor data (video, audio, and inertial measurement). These tests were selected because their results have been been shown to be predictive of MCI. We will implement phenotype extraction pipelines that employ advanced signal processing, machine learning, and computer vision algorithms to extract the target phenotypes from the sensor data collected for these tests and demonstrate they operate with sufficient accuracy to replicate published experimental designs. Successful completion of this project will eliminate the need for expensive and cumbersome phenotype collection equipment (e.g., eye tracking stations) and create the possibility of generating data from which MCI onset can be predicted. Data collected in Phase II via these and other such tests will enable us to apply our machine learning expertise to produce models able to predict transition to MCI that are both sensitive and specific, transforming any smartphone into an MCI risk assessment tool available for at-home use by millions of people.