Tech Talk: What is Artificial Intelligence, and how is it changing healthcare?

 

With the increasing popularity of using technology to increase efficiency, paper-based systems in hospitals are slowly being replaced by computers and electronic alternatives, only to find positive results. According to multiple studies, the use of electronic records instead of paper-based records for prescriptions can cut prescription errors by up to 80% [1] [2]. That’s a massive improvement, and it makes sense. Paper-based records can be misread, damaged or lost; and are not as easily accessible as electronic records.

As a result of using computers instead of paper, we now have a massive set of electronic data that encompasses the patient’s medical history, medication and prescription details as well as symptom and disease information. We can sort through and play with this data to detect existing patterns and discover anomalies. For a long time, we humans have been turning big datasets into visual representations such as charts and graphs and analysing them ourselves, but today we’re using artificial intelligence (AI) to detect and inform us of things we may have not noticed before and to do things that are much more difficult to do as humans. At the Medic App we’re looking at using AI in medication scheduling, but before we get into that let’s start with the basics.

 

What is Artificial Intelligence?

The Oxford Dictionary defines AI as “the theory and development of computer systems able to perform tasks normally requiring human intelligence”. In simpler terms, it is giving computers human abilities such as detecting objects in a photo, or recognising speech. Artificial intelligence also gives computers the ability to make a decision based on reasoning, as opposed to only following a flowchart of defined scenarios.

The concept of intelligent machinery was born centuries ago and existed in Greek mythologies such as Hephaestus’ creations. Since the dawn of time, as technology has progressed, humans have debated the thought of AI becoming “superhuman”. This debate still exists today, with some of the world’s brightest minds contributing to the topic and its use for military purposes [3]. But that’s a blog post for another time.

 

The Problem

We know (and have discussed in previous blog posts) that one of the largest problems in healthcare is the incorrect consumption of medication. It’s estimated that half of patients that are prescribed for chronic conditions do not consume their medication correctly.

Most medication is prescribed in a way that allows the patient to follow a standard routine. This is done in order to help the patient stay compliant and at the same time to consume the medication in a way that does not endanger the patient. For example, paracetamol, which is usually given in doses of 500mg per pill, is usually prescribed to be taken every four to six hours with a maximum consumption of 4mg per day. This allows the amount of paracetamol in the bloodstream to decay to a point where, when the patient consumes their next dose of the drug, there is still not enough of it in the blood to cause paracetamol toxicity. The same logic is applied to most painkillers including morphine and ibuprofen.

 

Using AI to Personalise Medication

Let’s make an assumption that a patient, Alice, takes an antibiotic four times a day. One of the requirements of this antibiotic is for it to be taken on an empty stomach. Knowing this, Alice has to also plan what time she should eat as well. What if, knowing what time Alice wakes up and sleeps, we can use artificial intelligence to actually suggest a meal and medication schedule for Alice?

We can also help Alice plan the times at which she wants to do other activities, not just her meals. Let’s imagine a different scenario where Alice is on painkillers. If she takes her medicines at 1pm, when should she go out to get groceries? We can also do this the other way: if Alice wants to go get groceries at 5pm, when should she take her medicines throughout the day while meeting that medicine’s constraints, in order for it to have the most impact?

What makes this an interesting problem is the fact that we have two conflicting objectives. The first objective is to minimise the amount of the drug that Alice would consume. If we know the maximum amount of paracetamol that Alice is allowed to take is 4g a day, then we obviously want Alice to take the least amount of the medicine as possible. The second objective is to maximise the relief from the medicine at the times which Alice needs the medicines to have the most impact.

This is the kind of problem you’d use AI for: solutions that would take too long to discover on paper, or too difficult to compute by humans.

But personalising medicine isn’t the only thing that we can do with AI in healthcare.

 

Using AI to Diagnose & Treat

DeepMind [4], an AI company, was recently acquired by Google. They’ve partnered with the NHS to create a powerful platform to diagnose diseases and recommend treatments.

An astonishing 2.5 million research papers are published every year, many of which are medical papers [5]. No human can possibly read all those papers and learn their results, but a computer can. This is what DeepMind does: reads, learns and actually understands each study (as well as look at the diagnoses and treatments offered to patients in the past) to diagnose illnesses and offer treatments better than any knowledgeable doctor could.

 

AI is Pretty Cool!

Yes it is! It’s why I’m doing my PhD in AI. If you’re interested in knowing how AI is changing our lives today, don’t stop here! Check out CGP Grey’s video called Humans Need Not Apply.

By Fares Alaboud

CEO and Founder

References

[1] http://www.iis.sinica.edu.tw/page/library/LIB/TechReport/tr2008/tr08.html

[2] http://www.iis.sinica.edu.tw/papers/janeliu/8683-F.pdf

[3] http://www.newsweek.com/ai-asilomar-principles-artificial-intelligence-elon-musk-550525

[4] https://deepmind.com/applied/deepmind-health/

[5] http://www.stm-assoc.org/2015_02_20_STM_Report_2015.pdf