Skip main navigation

New offer! Get 30% off your first 2 months of Unlimited Monthly. Start your subscription for just £29.99 £19.99. New subscribers only. T&Cs apply

Find out more

Semantic Data for Artificial Intelligence

Hi, my name is Adam. And I’m going to talk to you about semantic data for artificial intelligence. Well, for artificial intelligence to work it must be filled with quality data. The thing is healthcare is data intensive but information poor to define this properly, data represents unorganized and unprocessed facts whereas information or aggregation of processed data which makes decision-making easier. This is not quality data. I have a diagram here that illustrate in a simplistic manner. The continuum of care for the healthcare sector starting from prevention and wellness to palliative care. And you can see that in the primary care sector the acute care sector as well as the community extended care sector.
All of them have the data view of gender and what is obvious is what they all have the category of gender. The classification of the data points is drastically different between the different institution within the continuum of care. This makes it impossible for health care institutions like this to generate meaningful insights from the data. The thing is this example is something as simple as gender. Imagine if you are trying to do artificial intelligence in the world medication, in the world of other care. This makes it literally impossible to make meaningful use of a data. To talk about topic further, I need to introduce to you something what we call interoperability.
Now interoperability is defined as the ability of two or more systems or components
to number one: exchange information
and number two: use the information that has been exchanged. In the world of health informatics, we have holistically speaking, three levels of interoperability. We start off very physical interoperability which describes the medium of connectivity, or the physical connection. In other words, it is really something to carry your data, this could be your network the thumb drive or even something like due to. The next level we have will be functional interoperability. Now functional interoperability is syntactic by nature. What this means is syntax referring to the structure of the communication. This is somewhat similar to the spelling and grammar rules we use in languages. and lastly we have semantic interoperability. Now semantic interoperability refers to the meaning of the communication.
This is the vocabulary dictionary or toss words, so if we take this setting into consideration, the physical interoperability is the monitor in front of you when I’m able to actually transmit my video to you what delivering the structure. The functional interoperability is the fact that I’m using English in the language that you understand. So how about semantic interoperability? I’d like to give you an example to illustrate this concept a little more. So, take for example, if I will give you the English word “Apple” What would you think of? Most people would think of perhaps Apple, the company that produces iPhone. Other people may actually think of apple as the fruits.
This is a classical example of semantic interoperability, where the meaning of the word actually carries different definitions. So, the ability for you to understand me when I say “apple,” might differ in certain settings. So let’s bring this example to a healthcare environment. I have, on screen, an illustration of three hospitals. Hospital1, 2, and 3. And you can see the data field defin of diagnosis was similar between the three hospitals are slightly different. So for hospital 1, they have acute asthma suspected as one view under diagnosis. Whereas hospital 2, they have a granularity where the diagnosi is separated into several fields. So in this case asthma, acute, suspected.
In hospital 3, we have two fields: asthma and the status will be suspected. Now, when you’re reading this, you will be able to understand from the visual aspect and your brain is able to compute if you will. That this is actually describing the same thing. To a computer, it’s a three different description and this is what I mean by semantic interoperability where the meaning being convey is subjected to interpretation. What are the secondary effects? Well, in such situation and in fact this is a common situation. We will be unable to perform analytics or any form of artificial intelligence with the data collected. So we need to urn the data that we have into usable information, into something that’s semantic interoperable.
And I’d like to give another example, in medical literature. So how many ways are there to say heart attack in English? Now we’re not even talking about different languages. Something as simple as English. According to SNOMED-CT, which is the reference terminology used in clinical terms. They are actually, six ways to say heart attack. Again, for a trained clinician you will build… you will be able to see the six terms and understand that they’re actually talking about the same thing. But to computer, there are actually six different classification. Again, this is the issue of semantic interoperability but a meaning convey is open for debate.
In addition to defining certain data fields it is also important to be able to interpret the entire paragraph or not-written. Let me give you an example. I have five English word here it says “I like fast food delivery.” So it is a very simple sentence that you might say every day. The question is when I say “I like fast food delivery.” Do I mean I love delivery of food that is fast? or do I like fast food delivery? Again, very simple phrase, simply five words but the meaning convey is drastically different. And imagine this context put into the ball of Health and healthcare, where you have clinical analysists or physicians.
This opens up to different interpretations and will potentially affect the quality and safety of care. So to achieve interoperability, most specifically semantic interoperability we need standards profiles and terminologies. Now these are very important components in the world of health informatics as it enables interoperability between the disparate systems that we have implemented in the real world. Terminologies to illustrate further are special words or phrase that are used in a particular field. The technical or special terms use in business art science or special subject, in this case health and health care, it is the nomenclature as a field study. And to achieve semantic interoperability, we need to utilize two terminologies.
To achieve semantic interoperability, we need to utilize terminologies or more other standards. So like it or not, terminology drives everything in healthcare, every medical assessment, order, procedure, result, diagnosis etc. represented in the electronic medical record, uses coded medical terminologies. if these terminologies are not implemented properly the data within the EMR will not be very usable or not semantic in nature. if not implemented properly, the data within the EMR will not be very usable, or not semantic in nature. And this will cause problems for your AI solution. With that I end up the lecture. I hope you enjoy the talk. Have a nice day

Dr. Adam Chee is a convergence scientist working on various aspects of Health (& Care), Informatics, Innovation, Technologies and Business aspects of the ecosystem while serving as the people’s conduit. This video describes the quality of data needed for different algorithms of artificial intelligence.

This article is from the free online

Artificial Intelligence for Healthcare: Opportunities and Challenges

Created by
FutureLearn - Learning For Life

Reach your personal and professional goals

Unlock access to hundreds of expert online courses and degrees from top universities and educators to gain accredited qualifications and professional CV-building certificates.

Join over 18 million learners to launch, switch or build upon your career, all at your own pace, across a wide range of topic areas.

Start Learning now