Case Studies

Tuberculosis is the world's deadliest infectious disease, killing over 1.6M people every year, affecting developing countries the most. Chest X-rays — the most reliable way of diagnosing tuberculosis — has drawbacks which makes it a less than optimal choice:

• Chest X-ray machines are not portable and expensive, making them inaccessible in regions of the world with high tuberculosis risk
• Exposure to ionizing radiation during screening makes this method problematic when frequent tuberculosis testing is required

Looking to create a simple, yet effective method of tuberculosis diagnosing, our client approached us to help with the development of a mobile system for in-the-field screening using AI based on a recording of a patient’s respiration.

The screening system consists of multiple modules — an AI processing module and a desktop app — that connect several devices: a stethoscope for recording the patient's breathing pattern and a medical tablet for on-the-go screening results review.

Respiration Recording

The diagnostic process starts with a specialized stethoscope which records the patient's respiration. The stethoscope's firmware allows it to connect with a tablet and send the recording to an AI module for detection of abnormal breathing patterns.

Desktop App

Using a desktop application run on a tablet, a doctor can assess the patient's health condition on the go. The app also runs an AI model which analyzes the breathing recording in real time. The app is not cloud-based due to the need for the app to remain functional offline.

Core app features:

• Patient registration
• Signing of examination consent forms
• Aid in performing medical surveys and patient examinations
• Record patient's respiration
• Run an AI-powered analysis of the recording
• Sync data with cloud storage

The app is designed to run on top of all other windows and cannot be minimized, which helps to conceal the tablet as a medical device to prevent theft.

AI For Audio Analysis

The breathing pattern of a person sick with tuberculosis has certain markers, like weakened breathing, muffled or dull breathing sounds, dry and moist rales, all of which have a visual representation in the spectrogram. A neural network, trained to detect these markers, is able to detect tuberculosis signs in real time based on the spectrogram analysis and patient's medical survey data.

The AI system acts as an indispensable assistant to the doctor, helping them detect anomalies that might otherwise be ignored and provide a higher quality of medical services to those in need.

The system is now fully functional and has recently entered a beta testing stage.