HOW SYNDROME ANOMALY DETECTION CAN MAKE OUR WORLD A SAFER PLACE

The Problem

Current syndrome and disease outbreak detection solutions are disjointed. Existing outbreak detection technologies utilize only traditional incident analytics to provide syndromic surveillance, with limited inter-health team analysis, ultimately compromising the quality of health care.

The Initiative

Our GoodLabs Healthtech Studio and A.I. Engineering Labs set out to develop an early detection system for anomalous syndromes and potential disease outbreaks in real-time. The goal was to leverage the latest in recent Natural Language Processing advancements to effectively deliver an automated end-to-end A.I. based monitoring system.


The Solution

Our Deep Learning researchers work together with epidemiologists to design a full processing pipeline that will form the basis of a real-time infection surveillance system. 

  1. We first take the patient conversation from hospital triage, ambulance, health clinic, and e-health forums and use clinically trained BERT learning models to extract a patient’s symbols and other anonymous information right from the mobile application. 

  2. The information is then processed through another NLP model for normalization into a standard ICD10 representation. 

  3. The next step is for the SVM model to classify the ailment and utilize specially trained Autoencoders and statistical models to perform syndrome anomaly and outbreak detection.

The Result

By using state-of-the-art deep learning, edge processing, cloud-based data management, and analytics services, we have developed an early prototype that can monitor syndrome anomalies and outbreaks from numerous online e-health forums and provide early alerts to authorized parties including governmental departments and health institutions. Leading data indicators will enable collaboration amongst researchers around the world to better prepare us for the next pandemic.