Facilitating Electronic Medical Records with Deep Learning

December 17, 2020
Facilitating Electronic Medical Records with Deep Learning

Artificial neural networks open new horizons for the EMR-software market.

Digital record-keeping has become popular in many industries, including healthcare. Experts at the US National Cancer Institute define “electronic medical record” (EMR) as “an electronic (digital) collection of medical information about a person.”

It has become common to use such records since the beginning of the 2000s. The latest available data suggests that adoption of national electronic health record systems grew by 46% globally over 2011-2016. And it’s no wonder: instead of mining mountains of paper archives, now, a doctor can open electronic medical records in a matter of seconds.

Many companies offer software development solutions for these purposes. According to ResearchAndMarkets, the global market for electronic health records will reach $30B in 2020.

And the future of electronic medical records is even brighter. Business Insider Intelligence forecasts that by 2025, all US physicians will use some system designed to accumulate data about their patients in digital form. By 2025, the global market will grow to $40 billion.

Image by Proxet. Forecast: Penetration of Electronic Health Record Systems in the US
Forecast: Penetration of Electronic Health Record Systems in the US

Another trend is the use of machine learning and Artificial Intelligence in healthcare. These technologies assist doctors with early detection of diseases, diagnosing complex health issues, and coordinating care plans. Additionally, they are used for drug research and discovery and creating life-like simulations for training purposes.

The use of machine learning technologies might mean a new step in the evolution of electronic health record-keeping. As accumulating information becomes easier and organizations get more insights with new analytical tools, electronic health records will become more valuable.

“The past decade has seen an explosion in the amount of digital information stored in electronic health records (EHR). While primarily designed to archive patient information and perform administrative healthcare tasks like billing, many researchers have found secondary use of these records for various clinical informatics applications. Over the same period, the machine learning community has seen widespread advances in the field of deep learning.”

Benjamin Shickel, Patrick J. Tighe, Azra Bihorac, and Parisa Rashidi, University of Florida in their article.

Functions and Advantages of a Digital Medical Record

Both healthcare professionals and their patients benefit from the use of electronic medical records:

  • Charting. The medical history of a patient is automatically recorded in a secure, digital space. A tangle of multiple paper charts and notoriously messy doctors’ handwriting become a thing of the past;
  • E-Prescribing. Pharmacies get prescriptions almost instantly. Physicians receive notifications about dosage, allergies, and potential drug interactions;
  • Patient Portals. Patients can access their medical information directly. Most EMR systems provide them with information about medication history, past visits, and lab results from a healthcare institution;
  • Decision Support: Patients are alerted and reminded about scheduled treatments and  recommendations for particular procedures;
  • Order Entry. Providers can store and send orders for lab tests easily. The orders are less likely to get lost in a pile of paperwork. Plus, errors are filtered out of the system before they can cause any trouble;

For healthcare providers, introducing an EMR system can result in an automated and streamlined workflow and less expenditures. The study, published in the American Journal of Medicine, reported that savings might primarily come from cutting drug expenditures, better capture of charges, enhancement in the use of radiology tests, and decrease in billing errors.

The patients receive an improved healthcare experience. They have full access to their medical history, which is important to get them and their families more involved in the process. The patients can also manage their chronic diseases more easily. The turn-around time for messages to a physician gets shorter, and patients receive more educational materials.

“Connecting patients with their personal health data is not only the right thing to do, but it also fosters personal responsibility and self-determination that are crucial to patients’ maintaining their health and controlling healthcare costs.”

Dan Mendelson, Founder and former CEO of Avalere Health

Types of Electronic Medical Records

The term “electronic medical records” and “electronic health records” (EHR) are often used interchangeably. Both refer to a digital version of a patient’s chart. But there are some differences between them. An EMR provides information about diagnoses and treatments by a particular healthcare specialist. EHR is a more comprehensive report describing overall health. Some other differences include:

  • Healthcare providers mainly use an EMR for diagnosis and treatment;
  • EMRs are used within an individual practice;
  • EHRs allow healthcare organizations to share a patient’s information;
  • EHRs make patient medical data accessible to different specialists, emergency rooms, imaging facilities, labs, and pharmacies. State lines do not bound them.

Keeping and Utilizing Medical Records with Machine Learning

Machine learning and deep learning technologies (artificial neural networks) gather and analyze electronic medical record data more efficiently than legacy methods. Technology improves digitizing of paper charts and interoperability, i.e. exchange and use of information between computer systems. Computer vision, combined with natural language processing, can be used to convert paper records and notes into an electronic form. With the tedious heavy lifting handled by computers, healthcare professionals can then review the records and adjust incorrectly transcribed information.

Additionally, these technologies can facilitate automated receipt of medical information directly from patients.

“We’ve developed several custom software development solutions for healthcare. Our medical triage software for hospitals has the option of adding an AI-powered, patient-facing voice attendant. It takes patients through the same process as a nurse with the help of voice prompts, and patients verbally respond to questions.”

Vlad Medvedovsky at Proxet, custom software development solutions company

A major benefit of ML is its ability to enrich EMR systems with analytical capabilities. Software can store and recall medical information quickly while providing healthcare specialists additional insights. Medical data mining may be divided into several broad categories:

  • Diagnostic Analytics aims at determining why a health outcome happened. For example, New York-based Prognos uses ML algorithms to identify patients requiring enhanced treatment options, thereby improving risk management and quality of operations;
  • Prescriptive Analytics is a comprehensive analysis of a patient’s data, designed to improve patient management quality. Prescriptive Analytics can be applied to individual cases or towards coordinating task flow among medical personnel. For example, CareSkore provides software to alert professionals about factors that may complicate care, such as infections, medication adherence issues, etc.
  • Predictive Analytics analyzes patient data to determine possible outcomes. It might be the probability of improving or worsening health conditions, or the likelihood of heritable disorders. For instance, Seattle-based startup KenSci analyzes patient health records using machine learning to predict patient risks.

A group of researchers from Deakin University Geelong, Australia created DeepCare, a vivid example of using deep learning in medicine for prediction. DeepCare is an end-to-end deep neural network that uses EMRs to predict medical outcomes. The tool can help model disease progression, recommend interventions, and warn about future risks.

How to Train a Deep Learning System

Deep learning is part of machine learning methods based on artificial neural networks (ANN). An artificial neural network is a system of calculations and feedback loops that allows a computer to mimic a human brain.

Three main parts of an ANN are:

  • Input layer
  • Hidden layer(s) (with neurons)
  • Output layer

The input layer uses an activation function to signal the neuron layer. Neurons on the neuron layer can then either be activated or not, depending on the weights assigned to the various neurons. If activated, a signal is passed to the next layer within the neural network. “Deep learning” refers to multiple layers within the network. The signal passes through all the hidden layers, and you receive an output (prediction).

If ANN didn’t make the correct prediction, this failure is fed back into the system using a cost function. The cost function is a calculation indicating the difference between the output value and the correct output. In response to the cost function, the ANN adjusts weights on the neurons. The model is then relaunched, and the cycle is repeated until the predictions become accurate.

The basic steps of building an artificial neural network are:

  • Preprocess data: remove outliers, inaccurate or missing values;
  • Import functions: you don’t have to write all the code from scratch; there are many popular neural network libraries; TensorFlow and Keras are among the most popular;
  • Create the layers: select function, which applies weighting, and the number of layers;
  • Fit the model: allow the ANN to adjust itself to the data within your configurations' limits.
Image by Proxet. Structure of an Artificial Neural Networks (ANN)
Structure of an Artificial Neural Networks (ANN)

This overview may provide you with a general idea about deep learning. But to build a commercially successful solution using such complex technologies, you will need professional support.

Proxet will provide you such assistance. We’ve used Artificial Intelligence and machine learning to design solutions for various industries—from self-driving cars in smart transportation to healthcare chatbots. Deep Recurrent Neural Networks and Deep Convolutional Neural Networks (ConvNet or CNN) are in our technological arsenal. We will arm your business with the latest technologies tailored to your goals and objectives.

Related Posts

No items found.