This page describes how the Hyfe Research application works and what features it has. It provides instructions on how to use the app and describes the use cases for the person that administers end-users and accounts for them (let's call this person “Admin”).
Most of the devices will require an additional setup. It’s required to disable battery restrictions for the Hyfe App as described in the Don’t kill my app! device page.
Only the Pixel devices, AOSP devices, or devices with Don’t kill my app! rating less than 2 thumbs down will not require any additional setups.
The first step for an Admin will be to receive from Hyfe a document with user credentials. Credentials consist of a series of (a) “Hyfe IDs” and (b) associated pin numbers. An example of the file format can be found here.
Credentials are important for multiple reasons. First, it serves as a way for both Hyfe and Admin to be able to communicate and visualize data on a participant/user, without needing to share any information about that person’s true identity, thereby ensuring privacy. Second, it allows linkage between a research “cohort” or pool of patients and each individual. Finally, the secret pin prevents data contamination by ensuring that nobody else can capture data using the ID in question.
If you have not received credentials yet, please submit the Cohort Creation Form.
Admin needs to install the app on the device. To do so an Admin has to open the Play Store application then tap on the search field and type Hyfe Research.
It is possible to open app page directly by this link Hyfe Research on the Play Store
Make sure to download Hyfe Research and not Hyfe Cough Tracker
After clicking on the search button on keyboard Play Store will show a list of applications. Admin needs to click on the Hyfe Research app from the list. The Play Store will show a detailed page of the Hyfe Research application. Admin can click on the Install button and installation will start automatically.
Once the Hyfe Research app is successfully installed, the Admin can run it for the first time.
The app will show a Disclaimer and User Agreement with links to Term of Use and Privacy Policy. Admin must accept them before going further.
More information how the app works can be found here - How it works.
Once Admin accepts Term of Use and Privacy Policy the app will prompt Admin to the Sign in screen. Here Admin has to enter credentials provided by Hyfe.
The sign in screen has 3 fields and Login button.
Important: If your study requires informed consent, or if you are subject to certain rules and regulations regarding capturing sound data for health, it is your responsibility to ensure that you are in compliance.
After successful sign in, microphone and location permissions will be prompted. For the app to work, permissions must be granted: click "allow".
Next, you will be brought to the "Calibration" page. Make sure that you have an internet connection, and have the participant cough when prompted.
Important: Ensure that the participant is coughing in a way that is in compliance with your institution's infection control policies.
The participants, not the Admin, must do calibration themselves. But in the case of the participant not being present, the Admin could grant the microphone permission and then close the app. When the participant later opens the app, she/he will be able to finish onboarding.
To finish the calibration the user will have to tap on the Recording Button (Hyfe logo on bottom of the screen). After the click application will show 3..2..1 counter and the participant is asked to cough only once right after the number 1 disappears.
After that the app will process the cough for a few seconds. If the app recognizes a cough it will say that cough was detected and ask to do it two more times. Otherwise the app will say that cough can’t be recognized and the participant will be asked to try again.
Causes for a cough not being detected might be:
Once the calibration is finished the app will always show the main recording screen. This screen just shows that recording is active. The app works in background all the time. But users should be warned to NOT to force “kill” the app. The app will show the running notification all the time while it’s working.
If participants don’t observe the notification in the notifications bar they should just find the app in their apps list and open it. After that they can go back or hide the app.
On the Hyfe Research app, the way to stop tracking or end a recording session is simply to log out. Logging out uses the same credentials as logging in, and requires an internet connection.
To log out from the app the Admin must click on the Logout button on the main screen. The app will show current user Hyfe Id and ask to enter a Pin Code for that user id. If the pin code is correct the app will log out the user and show the very first Hyfe screen.
Thereafter, the phone is ready to be used again for another study participant (ie, the cycle starts from Step 3).
General indications:
Hyfe recommends two options to carry the phone:
The Hyfe Dashboard (https://dashboard.hyfe.ai/) gives researchers full access to their data and provides them with the tools to test Hyfe’s performance for themselves.
Researchers may test Hyfe performance however they see fit. As a starting point, here we describe (1) a generalized field test workflow and (2) specific instructions for the most common test procedures among our global network of research partners.
Use this basic workflow for completing whatever specific field test that you choose to carry out.
These are the most common methods for evaluating Hyfe performance. These tests are listed in increasing order of complexity; we recommend starting with Test 1 and working your way up as you see fit.
Assess baseline performance in ideal conditions
Hold the phone close to your mouth (10cm - 30cm). Cough loudly 30 times, 20 seconds apart (for a total of 9 minutes of coughing).
Assess the impact of loud ambient noise on Hyfe performance
First, establish a baseline expectation (ideal conditions) using Test 1. Then conduct the Test 1 protocol in a noisy environment (e.g., a busy office space, an outdoor cafeteria, a pedestrian street, a city park, etc.).
Compare Hyfe performance across multiple devices
Log in to two or more devices using separate HyfeIDs. Complete Test 1 a - b, making sure that both devices are held an equal distance from your mouth.
Determine the effect of a carrying device on Hyfe performance
Carry out the Test 3 protocol, placing one device inside your carrying device (e.g., a necklace holder, arm band, waist bag, pocket, or purse). Use the second device as a “control” (not inside a carrying device). Make sure to note which device is used as the control.
Determine how far the phone can be from the user.
Carry out a truncated version of Test 1 at multiple distances from the phone:
We recommend using two devices in this test: one device, the ‘test phone’ is the one placed on a table or shelf; the second device, the ‘control’, is carried by the participant and always held close to the mouth while coughing. Be sure to note which phone is the ‘test’ and which is the ‘control’.
Contact us here research@hyfe.ai. To ensure that we are able to assist you as quickly as possible, please be sure to provide the following information.
Please allow up to one week for a response.
The dashboard is a web application for researchers and physicians to monitor coughing and device activity remotely and in real-time. It is a developmental technology which is currently undergoing rapid changes.
Credentials for users to access the Dashboard are created when the Admin submits the Cohort Creation Form. If you have not given access, please, notify (from the email address of the PI or authorized delegate) Hyfe of the email(s) that you’d like to use to access the dashboard (research@hyfe.ai). You’ll then be approved as a user and linked with your study cohort.
Please note that exist two types of credentials to access the dashboard:
Visit the page at https://dashboard.hyfe.ai/ and log in.
Once logged in, you’ll be able to select (drop-down menu) all of the cohorts for which you are designated as an authorized viewer.
Click on a cohort and select “retrieve data for this cohort”. On the upper-left panel you will find:
Hyfe uses a “deep learning” model to categorize explosive sounds as either “cough” or “non-cough”. This classifier assigns each sound a value between 0 (almost certainly not a cough) to 1 (almost certainly a cough). Hyfe assigns a threshold / cut-off, above which sounds are considered to be “coughs” and below which sounds are considered to be “non-coughs”.
The optimal point for clinical monitoring has been determined to be 0.85. Though this results in some false negatives (ie, a sound scored at 0.75 may be a cough and is incorrectly classified as non-cough given the high threshold), 0.85 has been identified as the optimal point wherein sensitivity remains high (>90% of all human-confirmed coughs have a score of 0.85 or higher), but specificity is sufficiently high so as to keep false positives acceptably low for the purpose of monitoring.
This means you were logged off. To log-in again, you can do a refresh and log-in again using your credentials.
You can log in the app, cough a few times, log-in the dashboard, click on the Hyfe ID you’ve used, click ‘Individual data’, scroll down and playback cough sounds. In the playback menu, you will also see two types of visualizations: a wave form chart, and a spectogram. A wave form chart can be used to establish if sounds are over-saturated (ie, “too loud”, perhaps due to the microphone being extremely close to the participant’s mouth); the spectogram can be used to see the frequency bandwidth (which is useful for identifying if a microphone has any frequency filters).
The threshold is set by default at 0.85, since it is considered to be a value that indicates with a very high probability that the recorded sound is actually a cough. Sounds with more than 0.5 are considered to be probable coughs as well, however it increases the likelihood of listening to “false positives”, which are sounds that do not correspond to coughs.
“Time granularity" refers to the XXX in the phrase "number of coughs per XXX"
The date_time field will show the earliest moment in each month/day/hour. It always rounds down to the "floor" of the period selected (ie, all coughs from 11:00 to 11:59 show up as 11:00 if picking hour).
You need to go to ‘Individual data’, select the user’s ID, select ‘Day’ in the time granularity scroll-down menu, select ‘Table’ and ‘Summary user table’. If you want the total number of coughs per hour, week or month you need to change time granularity accordingly.
Left.
Data will be uploaded in the cloud and shown in the dashboard if the phone has network/internet connection. However, until network connectivity is established, data won’t be displayed. The fact that data is not displayed in the dashboard yet does not mean that it has not been captured (ie, data can be saved to the phone for several days prior to uploading if no internet is available).
Coughs registered during the longitudinal monitoring period after onboarding.
No. They are included in the ‘Summary User Table’ and ‘Full User Table’.
Elicited coughs registered using the Solicit Cough app feature, which is only available after onboarding has taken place.
Yes, you can configure the number for a research cohort. The default number is 3 and you can increase it up to 10. To request this change you need to send an email to lola.j@hyfe.ai.
Onboarding coughs are the initial coughs collected from a participant at enrollment, to complete calibration.
Yes. ‘Onboarding’ and ‘non-onboarding elicited’ coughs are both elicited coughs.
Participants for whom at least five minutes of longitudinal tracking have been registered.
Onboarded participants are study participants that have completed the initial (solicited) coughs at enrollment.
A 0.5 second audio snippet captured by the app (coughs and non-coughs).
A signal, generated every ten minutes, to indicate that the app is actively tracking coughs longitudinally (used to determine tracked person-time).
They represent the same variable, which expresses the duration of the recorded session in seconds.
For example, in “2021-07-16T11:35:13Z”, the first chunk of figures indicate the day of the year (i.e. “2021-07-16” indicates the 16th of July of 2021” and the numbers that follow the T correspond to the local time in which the sound was recorded (i.e. at 11:35am for that particular timezone).
Note that if the recorded time does not correspond to the actual time, we may need to adjust the time zone of your cohort. Please send an email to lola.j@hyfe.ai
It’s a continuous numerical value that ranges from 0 to 1, both included. Values of 0.85 or higher are considered to have a very high probability of being an actual cough, whereas values of 0.5 or higher are considered to have just a high probability.
If TRUE, it indicates that the sound is a “Non-Onboarding elicited cough”.
If FALSE it can be an “Onboarding cough”, if the former variable is TRUE, or a “Longitudinal cough” if the former variable is FALSE.
If TRUE, it indicates that the sound corresponds to either an ‘Onboarding cough’ or to an ‘Non-Onboarding elicited cough’’.
If FALSE, it indicates that it’s a “Longitudinal cough”.
The app will send data to Hyfe’s cloud server automatically whenever the phone has network connection, whether it’s WiFi or cell phone data.
To ensure all data is uploaded in the server, admin must log-in with the Hyfe ID and associated pin, click ‘Statistics’ and check if there are any unprocessed sounds. If there are unprocessed data click ‘Unprocessed sounds’ (see screen below).
No matter which Hyfe ID is currently logged in, the app will show the total count of all recorded sounds stored on the current device from different Hyfe IDs.
One sound takes about 1-1.5 seconds to get uploaded, with internet capacity of 90mb/s.
If the phone has been without service or WiFi for several days, it can take minutes to hours to upload all the app data.
Once data is uploaded to the server, it will automatically be deleted from the app. If data remains in the app after upload, please inform the Hyfe team to help you solve the issue (lola.j@hyfe.ai).
Yes. However, if the phone device has been without service or WiFi for several days, we recommend you to wait five minutes and check that data is being uploaded correctly before logging out. The following screen will show if data is being uploaded:
Please check the screens for data upload in different scenarios:
1. Uploading started while logged in
2. Uploading in progress while logged out
3. Uploading in progress no matter logged in/out
4. There are unprocessed sounds while logged out. Uploading is pending and should start automatically in up to 20 minutes or the user can start it by clicking "Process now" button.
Hyfe Research App version: ar1.4.0. In order to use advanced features of Hyfe Research application, user should be logged in as Admin and use app version ar1.4.0: To log in as Admin click on the Admin Tab on the bottom of the screen, enter credentials and click submit. After credentials are validated, you will be able to use the features below.
To view locations where Study participant coughed, Researchers should open Cough map from Admin menu. In case there were no coughs that day, application will show empty view with notification.
Researchers may also view an information about sessions, their start and end time by selecting Sessions option from Admin page.
Some research study protocols require elicited coughs at follow-up periods. To record cough Researcher should click on Solicit Coughs option on the menu.
In opened screen Researcher clicks on Hyfe-logo button at the bottom of the screen. After small countdown Study participant should cough to record the sound.
When cough is successfully detected Researcher will see screen with confirmation. Researcher may record one more cough by clicking Capture another cough button or return to Admin menu by clicking Back button.
In case no cough or sound (Study participant didn't cough) is detected Researcher will see appropriate error message and may click Hyfe-logo button to capture cough again.
This feature is used to verify the high performance of the Hyfe Research App. By logging coughs throughout the day, Users will record precise times at which coughs occurred. Researchers can then compare that log to the cough detections registered by the app.
Researcher can enable/disable 'Log cough' feature for Study participants from the Admin page by enabling toggle button next to Log cough option in menu.
In case 'Log cough’ is enabled, Study participants will be able to see and use Log cough button on recording screen. Every time Study participant clicks Log cough button it will create a time record in Logged time of coughs table. Researchers may access this table from Admin menu by selecting Logged time table option.
Researchers may access this table from Admin menu by selecting Logged time table option.
To update the App, Admin has to open the Play Store application then tap on the search field and type Hyfe Research. Click the Update button and then Open button.
