Natural language processing (NLP for short) is a field of artificial intelligence that uses algorithms to understand and respond to human speech.
While we as humans perceive speech as something natural and unimpressive, for a long time it was impossible for machines to interpret and mimic it. This, however, has been rapidly changing. Today, NLP-based technology is everywhere. Amazon Echo and Alexa are NLP-applications that are used by millions every day. Businesses all over the world employ chatbots to save their customers from possible human interactions and to save their own resources by hiring less customer support employees. Google uses NLP to find the correct results, and Gmail uses it to predict the next words in your sentence. The list goes on and never ends.
TechCrunch points out that NLP has made more progress in the past three years than any other field in machine learning. New network architecture that was proposed in 2017 and improved later changed the NLP forever, making it much more powerful, accessible and applicable in various fields.
Healthcare is one of those fields. More and more people already face healthcare-related NLP through chatbots or their speech assistant. But of course, there’s so much more NLP can do for this industry. Before we dive into NLP applications in healthcare, let’s take a peek into the magic behind natural language processing. How exactly does it recognize and interpret human speech?
How does NLP work?
NLP works with the dataset, organizing it into subsets that make sense to the machine. For example, it breaks the text into semantic units. When the text is structured, the system applies algorithms in order to interpret every semantic unit (which can be a word or a group of words, such as “on the other hand”).
There are a number of possible algorithms, but the two most popular ones are rule-based systems and machine learning models. Rule-based systems interpret the text based on predefined grammatical rules. Machine learning models collect data all the time, learn from it, and make conclusions about the text based on pure statistics.
What are the NLP techniques?
There are many specific techniques in which NLP is applied across different fields, including but not limited to healthcare. Let’s see what some of them are.
Optical Character Recognition (OCR)
NLP is used to scan hand-written or printed text and convert it into a digital format. This is called optical character recognition, or OCR. This can be done with any documents, including unstructured data texts, tables and images. All these documents aren’t just digitized ― they are also “fed” to the machine as a set of data, which helps the machine analyze unstructured data and learn further.
Named Entity Recognition (NER)
Named Entity Recognition, or NER, is a process of entity chunking, entity extraction, and entity identification. Entity can be a product, location, organization, medical code, time expression, monetary value, or a person. NER locates and classifies any and all of those into predefined categories, such as Person, Company, Time, Location, etc. turning unstructured data into structured data.
Text classification (also sometimes called text categorization or information grouping) is used to analyze text data and assign labels or tags to different semantic units according to specific attributes: subject, document type, language, author, and so on. This is another way to make sense of large chunks of data.
NLP algorithms search for the required elements in the text so that you can find any specific words or phrases in the documents. They also search for synonyms and detect misspelled words. As we noted above, this is what Google uses in its services.
Sentiment analysis is a form of analysis that seeks to attain a specific tone (positive, negative, or neutral) to a data set. It’s widely used to gather real feedback about a product, company, service, or person from social media posts. This way, companies can understand what people really think about their product (or almost anything else) and add to the data they gather from more traditional feedback methods, such as focus groups, questionnaires, and interviews that are often not accurate due to social desirability bias and personal goals of those who provide feedback. This has been recently considered for medical purposes as well.
Named Entity Recognition (ASR)
NLP doesn’t stop at text analysis, although this is its main field of application. One might also need speech recognition and interpretation, and for that there’s a technique called Automated Speech Recognition, or ASR. This technique transcribes oral data into a stream of words using neural networks and hidden Markov models. The error rate is still high with ASR, and the technique requires further improvement.
These aren’t all methods used in neural language processing, but hopefully it’s enough to catch the sense of how NLP does its job. Now let’s look at the applications and see how this technology can improve the lives of healthcare professionals, add to the quality of healthcare services, help with medical research, and do so much more.
Application of Natural Language Processing in Healthcare
There are plenty of ways natural language processing can be applied in healthcare settings. We won’t be able to cover all of them because not all ideas have been tried or even proposed yet. And because there’s a word limit for this article. But we’ll go through the most common uses of NLP in healthcare and also through less common ways NLP has been or can be applied.
Improving clinical documentation
Currently, most information about patients all over the world is stored chaotically, on paper, in an unstructured form. This causes plenty of problems for healthcare facilities. The job of health professionals in every country includes tons and tons of paperwork, and every professional will passionately tell you that that’s not what they studied this long for. Clinical documentation is often lost or misplaced, and valuable data is lost with them. Precious time that could be spent treating people is wasted. You really can’t overestimate the necessity of NLP when it comes to clinical documentation. So how can it help?
The role of NLP in clinical documentation starts with an electronic health record (EHR). But this is just the beginning. NLP can use speech-to-text dictation or dubbing of the text for data entry of patient information into EMR (electronic medical records). With the electronic medical records that are categorized and analyzed with the help of NLP, doctors and nurses will be able to retrieve the information they need whenever they need it, using any gadget. Patients will be able to access it online, learn about their condition and track it. Test results will be received via email or messengers and accessible from anywhere.
An electronic system will also make it possible to access the patient’s information on a request from the government and researchers. Gathering statistics will, therefore, become much easier. This will also eliminate the paperwork involved in having insurance: in this case, a medical record and an insurance card become one. No one will ever again have to wait in line filling up pointless questionnaires while in pain.
Of course, this raises questions of data security, but security technologies are there to protect medical records in the same way they protect bank cards and other high-risk information.
Some hospitals have already applied NLP technologies on a local level and are waiting for the world to catch up. For example, Concord Hospital in New Hampshire, USA, applied Nuance’s Dragon technology as part of a move to Cerner’s Millennium EHR system. Concord clinicians can now dictate a patient’s information from any workstation or smartphone. Similar initiative has been introduced at Minneapolis-based Allina Health that also adopted Dragon transcription tools. The hospital reports that in December 2020 more than 1,550 providers and therapists were using NLP technologies, which saved about $250,000 in transcription costs that month alone.
Improving clinical decision-making
The use of NLP in healthcare, including for the improvement of clinical documentation, also helps the decision-making process of a clinician. With all the information available at hand, it’s easier to see the patient’s medical history, notice any irregularities, take allergies and other issues in consideration, assess the patient’s condition, and finally make an informed decision of the patient’s treatment. NLP is also being used to aid clinicians in checking symptoms and diagnosis, reducing the level of subjectivity and the possibility of medical errors.
Accelerating clinical trial matching
One of the reasons for the clinical trial failure to be as high as 80% is that it’s near impossible to find the right patients in time. If you’ve ever seen hundreds of posters and leaflets at any university looking for students with specific characteristics, you realize how hard it is.
Now let’s recall what NLP does best: analyze massive amounts of unstructured data. When it comes to medical data, the possibilities for clinical research are endless. One of them is discovering patients eligible for studies. This, firstly, will help to find people who might benefit from experimental treatment and, secondly, accelerate scientific discoveries.
Expanding scientific knowledge among researchers
Finding the right patients is not all NLP can do for science. It can also help expand knowledge. There are many works that are stored in an unstructured form. There are tons of unpublished papers, most of which are unpublished due to negative results. The lack of a platform where all scientific knowledge is collected results in studies being replicated time and time again, simply because researchers in, let’s say, Germany, don’t know that the English have already disproved this same hypothesis.
NLP could help with translating all this information into electronic format, organizing it, and analyzing it to create a full picture of research that’s been done on every disease.
Analysing how people talk about their health issues
Many conditions result in vastly different experiences for patients. This is perhaps most apparent in psychological conditions, such as autism, ADHD, schizophrenia, and so on. Our knowledge of these conditions is very limited. Often, people are reluctant to share information. Moreover, if you recall Doctor House, everybody lies.
NLP can add to our knowledge by analyzing the data related to the disease available and presenting it to us. That’s what has been done with ADHD: a Lexalytics data scientist used NLP to analyze data from Reddit, ADHD blogs, news websites, and scientific papers (PubMed and HubMed databases). Based on the output, they modeled the conversations to show how people talk about ADHD in their own words. Analyses like this could help us understand psychological and other conditions so much better.
Challenges for the adoption of NLP in healthcare
Despite the vast benefits of natural language processing, its mass adoption in healthcare is still a long way off. Surely, there are common sense reasons for that: fear of new technology, time, money. But also, NLP is not easy. There are hundreds of languages in the world, each with their own syntax. There are also special symbols, emojis, usernames, hyperlinks, all of which have to be taken into account when interpreting. But the biggest challenge of all is the existence of context. One word can mean different things depending on the context. For example, in medical vocab, the word “dermatome” can mean an area of the skin supplied by a specific nerve root or a surgical instrument used to cut the skin.
Thankfully, NLP in healthcare doesn’t have to deal with sarcasm, because in some other fields, such as market research, this is another huge challenge. But we hope doctors don’t joke too much in their clinical documents. Despite these challenges, it’s safe to say that NLP can already be extremely useful in healthcare, and it will only get more useful in the future.