Danso

Anthropometry Artificial Intelligence: Proof-of-concept

Client Request

DANSO is a medical devices company founded by a group of enterprising students at Columbia’s Graduate School of Public Health. Mobile health workers in low-resource environments such as rural Africa need to do anthropometry on children and infants in order to determine their nutritional status. Currently, they use cumbersome devices that health workers have difficulty carrying around with them. As a result, often times children are not measured, nutritional status is not determined, and appropriate healthcare is not delivered.

DANSO came to ReadyMadeGO looking to build a more user-friendly way of measuring children and infants in low-resource environments. They wanted a program that could, with no more than 5 strategically taken images of the subject, determine their height, head circumference, and mid upper arm circumference. ReadyMadeGO built a proof-of-concept application for them that used cutting edge artificial intelligence techniques and clever algorithmic design to demonstrate that this concept was feasible.

The Artificial Intelligence detects and localizes various features on the subject’s face.

Architecture

The first step of the work was determining an appropriate technical architecture. After conducting a thorough investigation of what techniques were appropriate for their use case, ReadyMadeGO settled on the following:

The user inputs various photos of the subject, which are then processed in the computer vision layer.

In the Computer Vision layer, various convolutional neural networks and vision APIs are applied to detect and localize key features on the subject.

In the Initial Calculation layer, proprietary algorithms are used to combine features and derive estimates for height, head circumference, and mid upper arm circumference.

In the Refinement W/Regressor layer, custom machine learning algorithms are applied to refine the initial calculation. Refined measurements are then reported.

The core of the algorithm uses cutting-edge computer vision techniques to localize and detect various features on the subject. We made heavy use of convolutional neural networks, and invented proprietary algorithms to calculate the subject’s height, head circumference, and mid upper arm circumference.

The Artificial Intelligence detects and localizes the subject’s entire body.

Technologies

CaffeFacebook Artificial IntelligenceGoogle Cloud PlatformGoogle Cloud VisionLinuxLuaPythonTorch

The artificial intelligence detects and localizes individual body parts on the subject.

The Artificial Intelligence detects and localizes various joints on the subject.

A boutique software development agency specializing in web applications