GOOGLE AI: PREDICT WHEN A PATIENT IS GOING TO DIE

Have you ever thought when are you going to die? When will you take your last breath? After thinking of these questions you might rush to a priest or a tarot card reader or someone whom you think can tell your future. But let me tell you neither you know your future nor those future teller do. The one who knows it accurately is the Google Artificial Intelligence.
Shocked?? Yes, you heard it right. Google is working on a project which will predict your death. And to our surprise it has also completed a couple of tests and found it to be 80% to 85% correct. This is the real artificial intelligence.
How this is being done? It’s all about deep learning for Electronic Health Record (EHR).
What is Electronic Health Record?
According to a blog published by Google AI on 18 may,2018 deep learning for Electronic Health Record is an application of machine learning. They say that when a patient is hospitalized it becomes necessary for both patient and doctor to know actual condition of the patient. The answer to this question helps doctors and nurses to take better care and the patient can be diagnosed efficiently. And if the doctors get to know that the patient’s health is deteriorating the treatment can be made faster.
So in this case deep learning for EHR comes into picture. The prediction of this deep learning will be useful for the doctors in many ways. For prediction to be useful it should be at least:
1. Scalable
2. Accurate

Now let us discuss these two in detail:
SCALABILITY
Electronic Health Records (EHR) is complicated to use as each hospital has its own way of representing the data of the patients. This makes data of one hospital look different from data of another hospital. Even the temperature measurement has different meaning and is based whether it is measure under the tongue, through eardrums or at your arm pits. So data like this is not scalable and difficult to use for prediction of patients health.
So for applying machine learning patient’s record must be represented in a consistent way. Google AI has an answer to this also. Google provides hospital with Fast Healthcare Interoperability Resources (FHIR) standards which is a consistent way to represent patient’s record.
Once a consistent format is there, now there is no need to enter the data manually to apply deep learning process on it to get the prediction. Instead from FHIR format the deep learning model reads all the data point from earlier to most recent and then gives the accurate prediction. But here is a limitation that there are thousands of data points. Now which one to use to predict the outcome? So to overcome this limitation Google AI will be developing the new deep learning model based on Recurrent Neural Networks (RNN) and feedforward networks.

Also Read:-

AI: ARTIFICIAL INTELLIGENCE AND ITS USE

So using FHIR the data is made in a consistent way but one thing which is still left is the accuracy. Now let us discuss on this.
Accuracy
According to Google AI the most common way to access accuracy is by a measure called area under the Receiver Operator Curve (ROC). This measures how well a model distinguishes between a patient who will have a particular future outcome compared to one who will not. According to this when the number is 1 means that model is prefect. So higher the number means more accurate the model is. According to the test of the model by Google AI it recorded following values in different cases.
1) Google AI model scored 0.86 in predicting if patient will stay longer in the hospital whereas traditional methods scored 0.76.
2) Google AI model scored 0.95 in predicting inpatient mortality whereas traditional methods scored 0.86.
3) Google AI model scored 0.77 in predicting unexpected readmissions after patients are discharged whereas traditional methods scored 0.70.
So what does this show? This clearly tells us that deep learning for Electronic Health Record using Machine Learning gives us a better accuracy than the traditional methods.
A proof of this concept is also given by the Google AI in their blog. They called this an “attention map”. The attention map is shown below, please refer it for clarifications.

At last just think of the future we are going to have. We are at the verge of technology. And this one is just the trial of Google AI and is just the beginning of the work. Google AI will be looking after all its limitations and is going to increase the accuracy of the prediction in the near future.
Hope you have liked the article. Please do comment below for your suggestion and questions.

ALSO READ:-

Play YouTube in the Background – Android : WITHOUT ANY SOFTWARE!

About Neeraj Kumar

If anyone asks whats the soul of this site - spit out dis single word - Neeraj! Whole damn success goes to this genius who is on facebook, insta, Google+, LinkedIn, WhatsApp....all at the same time! This Engineer in the making is quite popular for his funny one liners, and optimistic attitude.

View all posts by Neeraj Kumar →

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.