forked from resublime/DBDP
-
Notifications
You must be signed in to change notification settings - Fork 2
Digital Biomarker Discovery Resources
RunsData edited this page Nov 5, 2020
·
3 revisions
The Digital Biomarker Discovery process can be challenging. We have compiled some resources that, along with the DBDP, can be used to make digital biomarker discovery more robust. Please open an Issue if you have an idea for additional resources that we could provide here!
- Paper on wearables and the medical revolution
- Paper on quantifying the use of connected digital products in clinical research
- The Playbook: Digital Clinical Measures
One of the most FAQ when starting digital biomarker development is how to choose a sensor/device. One of the major considerations should be evaluating how accurate the wearable sensor is.
- In this study, we evaluated 6 of the most commonly used wrist-worn devices for accuracy.
- Elektra Atlas: a searchable catalog of 750+ biosensors with clear comparison charts based on validation, usability, security, data rights and more.
- The DBDP provides some data handling in the Pre-processing Module. The DBDP is currently being integrated with MD2K and Open mHealth data handling schemas.
- MD2K Cerebral Cortex provides an interface to retrieve/store mobile sensor raw data and metadata.
- Open mHealth Shimmer retrieves mHealth data and metadata and converts it to JSON
- V3 Framework for Verification, Analytical Validation, and Clinical Validation
- What is the problem you are trying to solve? Does it require domain expertise? Particular computational skills?
- How will you get your data? What file format will it be in? What is your desired data format? Do you have the computational tools needed to get the file into the desired format?
- In the DBDP, you will find many pre-processing functions and notebooks to help you pre-process your data from mHealth and wearables.
- Exploratory Data Analysis (EDA) is a critical step in developing digital biomarkers. Your EDA process will depend on your clinical problem, but we provide the basics in the Exploratory Data Analysis module!
- If you will be using traditional machine learning methods, you will need to engineer features. These can be data-driven or domain-driven.
- The DBDP provides several modules for feature engineering and are continuing to add more weekly! Current modules include cgmquantify for CGM data, wearablevar for wearables, and HRV for ECG/PPG. You will also find Feature Engineering tools in our other modules, including the Sleep Module, RHR Module, and the Human Activity Recognition Module.
- In the MLMethods Module, we provide machine learning methods
- Check other modules for ML methods, including the Sleep Module and the Human Activity Recognition Module (The HAR Module contains both traditional and deep learning methods)
- Digital Medicine Society (DiMe)
- Digital Medicine Primer
- Digital Medicine Education Tutorials from Elektra Labs
- DiMe Library of Digital Endpoints