Child Growth Monitor

(This repository is being updated further- For any up to date information, please reach out at info@childgrowthmonitor.org)

We provide quick and accurate data on child malnutrition in the most ethical way to frontline healthcare workers and organizations sharing our vision of Zero Hunger

Links

• Donate!
• Frequently Asked Questions
• Website

Table of Contents

<!-- TOC depthFrom:1 depthTo:3 withLinks:1 updateOnSave:1 orderedList:0 -->

• Problem
• Solution
  • Mobile App
  • App Backend
  • Machine Learning Backend
  • Machine Learning pipeline
• Data
• Scanning Process
• Impact

<!-- /TOC -->

Problem

Hunger or malnutrition is not simply the lack of food, it is usually a more complex health issue. Parents often don't know that their children are malnourished and take measures too late. Current standardized measurements done by aid organisations and governmental health workers are time consuming and expensive. Children are moving, accurate measurement, especially of height, is often not possible.

Bottom line: accurate data on the nutritional status of children is unreliable or non existent.

Solution

Our users do a quick scan of a child, similar to recording a video. We use the data from the smartphone camera and further sensors to measure the child using machine learning and artificial neural networks. Therefore we provide a quick and touchless way to measure children and detect early warning signs of malnutrition.

Mobile App

Github App Repository

The mobile app provides authenticated users an interface to scan children in 3D with consent of the parents and upload all collected data to the secure backend.

We guide the user to scan the child in a way that a quick, accurate measurement can be taken. This will involves data of the camera and child pose, depthmap and RGB images.

Because of the limitations of mobile connectivity in rural areas and in slums with tin roofs offline first is a major goal of the project. While the app already works fine in offline environments, results from the scans are currently produced in the cloud. Providing predictions directly on the device is the next big step we are taking, as it would also improve privacy by not having to upload every scan.

App Backend

Backend is implemented in Azure and uses

• Authentication with an Azure B2C tenant via OAuth2
• Flask API for backend
• Vue.js Frontend for data analysis and cleaning
• Custom Python ETL processes to anonymize data
• AzureML
• Storage Accounts are used with Queues and Blobs for structured data and scan artifacts
• PostgreSQL for structured data
• Grafana and Apache Superset for monitoring and evaluation dashboards

Database

PostgreSQL is used for structured data.

Storage

Storage Blobs are used for large objects such as rgb images and depthmaps. Storage Queues are used for transfering structured data from app to backend.

Machine Learning Backend

Development of the machine learning backend happens at Github ML repo and on our DevOps project.

Data

Please refer to our OpenAPI description.

Scanning process

Before any data is accessed or added our trained team explains to parents in simple terms that

• the data belongs to the children and parents,
• they give us the rights to store and process the data for a limited time
• this right can be revoked any time and the data deleted
• we are using the data only for the achieving the UN goal of Zero Hunger by 2030

Lastly, the informed consent with the caregivers signature is scanned to document compliance.

The scanning process is broken down into three parts for each standing and lying children. We evaluate scanning results to find the best way of scanning to gather necessary data. Children wear underwears while getting scanned.

Front scan

The child is scanned from the front.

Back Scan

The child is scanned from the back.

Side Scans - Right and Left

The child is scanned from the side left and side right.

Impact

The main advantages of our solution compared to traditional measurement methods are that the Child Growth Monitor:

• delivers accurate anthropometric measurements (close to minimum error threshold)
• enables anyone with supported smartphone to be an expert for anthropometric measurements
• is faster
• is easier to do and less stressful for everyone involved
• gets rid of unreliable, bulky and expensive hardware
• eliminates the possibility to manipulate data

Impact during the corona crisis

Measuring malnutrition had been stopped in many developing countries due to Covid-19. Traditionally, children need to be touched to be measured which was not possible during Covid-19. Our app offers a no-touch solution.