GyroGlove™
September 1, 2021Women’s Healthcare
September 1, 2021Babylon Health
Babylon's mission is to offer health services by creating an AI system that can analyze medical datasets and patient health records...
Overview
In their mission, Babylon Health wants to offer an accessible and affordable health service into everyone’s hands by combining the human experience and the power of Artificial Intelligence (AI). They built an AI system that can effectively receive, analyze, and learn from anonymized, aggregated, and consented medical datasets, patient health records, and the consultation notes that their clinicians make. The set of AI-powered tools and associated systems can break down this information to rapidly assist the decision-making process, including triage, causes of symptoms, and prediction of risk factors.
Babylon Health AI Aspects
What AI does
AI uses aggregated, anonymized, and consented data from various sources such as medical datasets. It can also read and learn from patient health records, like consultation notes made by clinicians, in each of their locations.
It can reason to decide on the likely causes of people’s symptoms, provide information about them, and recommend the following steps, including treatment information and disease risk based on the individual’s current health and behavior.
It can make predictions based on facts and observations. It simulates future potential outcomes to coach and nudges behavioral change to reduce the likelihood of future illness.
What AI does
AI uses aggregated, anonymized, and consented data from various sources such as medical datasets. It can also read and learn from patient health records, like consultation notes made by clinicians, in each of their locations.
It can reason to decide on the likely causes of people’s symptoms, provide information about them, and recommend the following steps, including treatment information and disease risk based on the individual’s current health and behavior.
It learns with every experience allowing it to gain more accuracy and interact in a more personalized manner over time. When a patient answers questions about their lifestyle and family history Health check will generate a report about his overall health. They explain it is like building a Digital Twin to explore the patient’s body, learn about medication use or family history. People will get recommendations to improve their health today and in the future.(1)
It can make predictions based on facts and observations. It simulates future potential outcomes to coach and nudges behavioral change to reduce the likelihood of future illness. It learns with every experience allowing it to gain more accuracy and interact in a more personalized manner over time. When a patient answers questions about their lifestyle and family history Health check will generate a report about his overall health. They explain it is like building a Digital Twin to explore the patient’s body, learn about medication use or family history. People will get recommendations to improve their health today and in the future.(1)
How A.I Works
AI revolves around four main parts:
Knowledge base:
AI works with a digital encyclopedia of medicine that contains the definitions, characteristics, and relationships of the different diseases, symptoms, and treatments. It objectifies this information with a graphical characterization that shows the relationships between the medical components.
AI revolves around four main parts:
Knowledge base:
AI works with a digital encyclopedia of medicine that contains the definitions, characteristics, and relationships of the different diseases, symptoms, and treatments. It objectifies this information with a graphical characterization that shows the relationships between the medical components.
Comprehensive health record:
The Health record is a collection of all relevant information related to the patient, including their medical history and private data put together through interactions with software. It helps to make connections between patients and multiple pathologies and their probable evolution over time.
Probabilistic graphical model:
The probabilistic graphical model uses the knowledge from the digital encyclopedia, combined with all the data, to test different models about illnesses. It is capable of processing millions of combinations of symptoms, diseases, and risk factors per second to help identify conditions that may match the information entered.
A similar approach can be applied to predict multiple risk factors over the next five years.
Simulations:
Simulations are used to estimate ‘what-if’ scenarios, to predict what happens if people continue their routines of diet, exercise, sleep, and stress. It helps users understand the impact of their actions and helps them develop optimized care plans.(1)
Babylon Health IT
Their intellectual property includes 18 registered patents primarily in the ‘Computing; Calculating’ category. Also, they have registered seven trademarks, with the most popular class being ‘Medical services; veterinary services.’ (2)
Babylon Health Financial Aspects
Since its foundation in 2013, Babylon has raised a total of $635.3M in funding; this includes $550 million in Series C funding. Saudi Arabia’s Public Investment Fund, an investment fund representing the Saudi Arabian government, led the round that increased the company’s value to $2 billion. It is currently funded by eight investors, of which Munich Re/ERGO Corporate, Venture Fund, and VNV Global are the most recent. The company’s current annual revenue is estimated at around $206.3 M per year.
The new influx of capital will enable the company to expand into other markets, including the US and Asia. Also, it will increase its AI capabilities that are currently serving more than four users around the world. (3)
Babylon Health Financial Aspects
Since its foundation in 2013, Babylon has raised a total of $635.3M in funding; this includes $550 million in Series C funding. Saudi Arabia’s Public Investment Fund, an investment fund representing the Saudi Arabian government, led the round that increased the company’s value to $2 billion. It is currently funded by eight investors, of which Munich Re/ERGO Corporate, Venture Fund, and VNV Global are the most recent. The company’s current annual revenue is estimated at around $206.3 M per year.
The new influx of capital will enable the company to expand into other markets, including the US and Asia. Also, it will increase its AI capabilities that are currently serving more than four users around the world. (3)
Predicting Stress Levels with the Use of AI
The coronavirus pandemic has increased stress levels in people across the US, and AI could help establish which cities are stressed the most.
A group of researchers from the UK is trying to answer that question using a Machine Learning system called TensiStrength. This system can detect stress levels in text, analyzing parameters like punctuation, word choice, and pronunciation.
The study processed more than 155,000 coronavirus-related tweets from the top five most populated cities for each state in the US through their system. The results showed Wyoming as the most affected state, with 40.1% of the tweets showing some level of stress. South Dakota was the least stressed, where only 19.4% of the tweets manifested some level of stress.
Stockton, California, was the worse ranked city, with 43.7% of the tweets manifesting stress about the disease; contrary to the lowest ranking East Honolulu, where only 16.5% of its coronavirus-related tweets registered as stressed. (4)
Predicting Stress Levels with the Use of AI
The coronavirus pandemic has increased stress levels in people across the US, and AI could help establish which cities are stressed the most.
A group of researchers from the UK is trying to answer that question using a Machine Learning system called TensiStrength. This system can detect stress levels in text, analyzing parameters like punctuation, word choice, and pronunciation. The study processed more than 155,000 coronavirus-related tweets from the top five most populated cities for each state in the US through their system.
The results showed Wyoming as the most affected state, with 40.1% of the tweets showing some level of stress. South Dakota was the least stressed, where only 19.4% of the tweets manifested some level of stress.
Covid-19 Care Assistant
One of the consequences of the COVID-19 pandemic was the development of mobile medical applications that assist in disseminating crucial medical information, checking symptoms, and training medical personnel.
Covid-19 Care Assistant is a mobile healthcare application developed by Babylon Health that empowers users to monitor their current symptoms and assists them with online medical consultations. Depending on the severity of the symptoms, it can also refer patients to specific hospital care.
The app uses an AI live chat service that inquires about concerning symptoms and offers guidance on the next steps. A specialized clinical team controls the chat under the supervision of doctors. This tool guides people in the form of six steps, starting with general information on the outbreak and advice on self-isolation, caring for others, and minimizing contamination. (5)
Covid-19 Care Assistant
One of the consequences of the COVID-19 pandemic was the development of mobile medical applications that assist in disseminating crucial medical information, checking symptoms, and training medical personnel.
Covid-19 Care Assistant is a mobile healthcare application developed by Babylon Health that empowers users to monitor their current symptoms and assists them with online medical consultations. Depending on the severity of the symptoms, it can also refer patients to specific hospital care.
Covid-19 Care Assistant
One of the consequences of the COVID-19 pandemic was the development of mobile medical applications that assist in disseminating crucial medical information, checking symptoms, and training medical personnel.
Covid-19 Care Assistant is a mobile healthcare application developed by Babylon Health that empowers users to monitor their current symptoms and assists them with online medical consultations. Depending on the severity of the symptoms, it can also refer patients to specific hospital care.
The app uses an AI live chat service that inquires about concerning symptoms and offers guidance on the next steps. A specialized clinical team controls the chat under the supervision of doctors. This tool guides people in the form of six steps, starting with general information on the outbreak and advice on self-isolation, caring for others, and minimizing contamination. (5)
Babylon Health on Learning Medical Triage from Clinicians Using deep Q-Learning
Medical triage is an efficient approach through which clinicians inquire about the causes of the patient presenting symptoms and can ultimately provide a plan for the following course of action.
Deep Reinforcement Learning aims to triage patients using clinical vignettes by implementing trained agents that learn when to stop asking questions or inquire more information in case of a new clinical scenario. Essential algorithms are used to learn policy functions and value functions directly. Compared to supervised methods, it has the same performance while demanding much less evidence. (6)
The Use of AI to Provide Accurate Medical Diagnosis
In the US alone, 5% of outpatients receive a wrong diagnosis every year, resulting in serious patient harm and even death. This represents a current challenge in the healthcare system. The use of AI in medical diagnosis is promising. However, until now, ML systems only provide advice using algorithms that rely on correlations – an approach that is helpful only for simple diagnosis.
That is why researchers from Babylon Health created a new AI capable of playing alternate scenarios and considering if the symptoms seen in a current case could be present in a different diagnosis scenario, allowing the system to set apart a potential diagnosis more efficiently.
They designed a study comparing the new AI’s diagnostic accuracy, a standard ML system, and a cohort of 44 doctors, using a test set of 1671 clinical vignettes.
The results showed that the doctors achieve an average diagnostic accuracy of 71.40%. In comparison, the old algorithm achieved a similar accuracy of 72.52%, placing it in the top 48 percentile of doctors in the cohort. However, the new algorithm achieved an average accuracy of 77.26%, placing it in the top 25 percentile of the cohort and achieving expert clinical accuracy, with even better results for rare diseases.
The data is encouraging, with the potential of speeding doctor’s diagnoses, improving accuracy and patient outcomes (7)
The Use of AI to Provide Accurate Medical Diagnosis
In the US alone, 5% of outpatients receive a wrong diagnosis every year, resulting in serious patient harm and even death. This represents a current challenge in the healthcare system. The use of AI in medical diagnosis is promising. However, until now, ML systems only provide advice using algorithms that rely on correlations – an approach that is helpful only for simple diagnosis.
That is why researchers from Babylon Health created a new AI capable of playing alternate scenarios and considering if the symptoms seen in a current case could be present in a different diagnosis scenario, allowing the system to set apart a potential diagnosis more efficiently.
They designed a study comparing the new AI’s diagnostic accuracy, a standard ML system, and a cohort of 44 doctors, using a test set of 1671 clinical vignettes.
The Use of AI to Provide Accurate Medical Diagnosis
In the US alone, 5% of outpatients receive a wrong diagnosis every year, resulting in serious patient harm and even death. This represents a current challenge in the healthcare system. The use of AI in medical diagnosis is promising. However, until now, ML systems only provide advice using algorithms that rely on correlations – an approach that is helpful only for simple diagnosis.
That is why researchers from Babylon Health created a new AI capable of playing alternate scenarios and considering if the symptoms seen in a current case could be present in a different diagnosis scenario, allowing the system to set apart a potential diagnosis more efficiently.
They designed a study comparing the new AI’s diagnostic accuracy, a standard ML system, and a cohort of 44 doctors, using a test set of 1671 clinical vignettes.
The results showed that the doctors achieve an average diagnostic accuracy of 71.40%. In comparison, the old algorithm achieved a similar accuracy of 72.52%, placing it in the top 48 percentile of doctors in the cohort. However, the new algorithm achieved an average accuracy of 77.26%, placing it in the top 25 percentile of the cohort and achieving expert clinical accuracy, with even better results for rare diseases.
The data is encouraging, with the potential of speeding doctor’s diagnoses, improving accuracy and patient outcomes. (7)