Can OCR and AI “Read” Medical Records?

The number one question I get from healthcare organizations is “Can Optical Character Recognition (OCR) “read” medical records?” I have been working with OCR for over 25 years and thought I would spend a couple of minutes sharing what I have learned (and what I am relearning) along the way.

The short answer is yes, OCR can be very effective in reading all kinds of healthcare documents, including medical records. It is equally important to understand how the technologies work and what are some of the challenges that remain when using OCR to read medical records.

What is Optical Character Recognition (OCR)?

OCR is a set of technologies that turn images, faxes, PDFs and other documents into data so that they don’t have to be keyed in by people. Why do these documents need to be turned into data? It might be helpful to use an analogy. If you were to take a photo of a document from your mobile phone, it is only an image of the document. None of the words on the page are searchable. OCR software analyzes that image and converts the picture of words into characters as if you had typed them yourself. Those converted characters are now useful data that can be sent to systems or stored with the document to make them more easily searchable in the future.

Why would I want to use OCR for medical records?

Despite the great work of standard committees, associations and government agencies, practically, we remain years away from being able to widely share medical records data between healthcare organizations. Today, most medical records are shared on paper (which must be scanned into digital images), via fax (which are transmitted as images) or uploaded to portals (both images and PDF documents). The images are great, but healthcare providers and payers need the data from those documents to effectively care for patients and manage the payment of medical claims. This is where OCR comes in. OCR can convert those images and PDF documents into data that can be usable by systems, people and workflows at multiple organizations. Importantly, OCR allows the medical record images to be searchable so that, for example, a nurse can search for all the lab results with the phrase “Liver Panel” in them and go right to the results.

OCR: The Rodney Dangerfield of Software Technologies

Many people I talk with are skeptical about OCR. I commonly hear concerns about OCR accuracy, complexity of setup and inability to read hand printed or unstructured documents. I applaud a healthy amount of skepticism about most technologies and OCR is no different. That said, OCR has been around for decades and has improved greatly over that time.

Traditional OCR

What I call “traditional OCR” was based on fixed templates that could only read nicely machine printed, highly structured forms, like the notorious red drop out CMS 1500 health claim form. Traditional OCR could not deal well with loosely structured documents like an invoice or an unstructured document such as a medical record. Fortunately, new technologies have emerged that are displacing older technologies and opening new opportunities.

OCR with Artificial Intelligence

Newer OCR technologies leverage Artificial Intelligence (AI) and Machine Learning (ML) to radically improve the success of OCR on highly unstructured documents like medical records. AI allows OCR software to adapt to a document’s context, searching for a piece of data that may not be in the same place on every form, such as a blood pressure reading. ML can “learn” via self-teaching and interactive training that allows it to learn different documents quickly, with little or no human intervention required.

What are the challenges with reading a medical record?

There are a number of challenges in reading a medical record with OCR. Quality varies greatly, especially fax quality. There are also millions of different document formats that need to be read. Handprinted characters and legibility of physician writing can also cause challenges. OCR is not easy. In fact, OCR of medical records with traditional OCR solutions is impractical at best. However, combining OCR, AI and ML changes the game. The adaptive, learning nature of the AI/ML combined with OCR overcomes most of the challenges listed above. It is now technically practical to read medical record documents and convert these images to usable, searchable data.

What kind of results can I expect?

Another common question I get is, “What kind of accuracy am I likely to achieve?” Simply stated, like miles per gallon on your car, actual mileage varies. Same with OCR. So, what have I learned in my 25 years of working on this problem?

Strategy over technology: Establishing expectations and a strategy is key. While many customers look to technology and tools first, it is more valuable to understand that success with OCR is an incremental journey, not an event that happens once you buy the right product. OCR requires a plan.
Show Me: It’s common for OCR software vendors to make many claims about their products. No problem. It’s quite simple to determine if the claims are true. Run a series of tests on real documents and show the results. At BRYJ, we do this as a part of every project and believe it is a critical step before purchase. Your vendor should not charge you to prove their product works on your documents.
Start Simply: OCR can be complex; your first projects should not be. Start with a simple initial project that creates a foundation for the future while delivering some meaningful return on investment. Your first project should take less than three months, or it is too big.

Getting started

Hopefully you now understand that OCR can read medical records with the advent of Artificial Intelligence and Machine Learning. I also discussed that strategy is more important that technology in most OCR projects. Having the help of an experienced partner can make a huge difference on your project. If you would like to discuss how BRYJ helps our customers develop and implement winning OCR solutions for medical records (and other documents), please contact us by filling our contact form.