Understanding AI Lingo in Healthcare

Share with your network

With the ever-growing incorporation of technology into medicine over the past decade, healthcare industries have advanced to integrate novel technology innovations. Such innovations include artificial intelligence (AI), virtual reality (VR), 3D-printing, robotics, and so on. One of these innovations, artificial intelligence (AI), holds promise in improving patient care while reducing costs. This technology has been applied in areas such as patient diagnosis and monitoring, treatment protocol development, radiology, and drug development. While some of this might seem like science-fiction, it’s being incorporated every day in the healthcare field. To help introduce you to this new world below, we’ve compiled a list of some of the most common terms in this field.

Basics of AI and machine learning

  • AI: “The study and design of intelligent agents.” In healthcare these agents gain information, process it, and provide specific output. Often healthcare AI programs in healthcare use pattern recognition and data analysis to evaluate the intersection of prevention or treatment and patient outcomes.

  • Algorithm: Instructions or rules for a computer to execute that solve problems or perform calculations. AI is dependent on algorithms to make calculations, process data, and automate reasoning.

  • Machine learning: Is a type of algorithm that “self teaches” or improves through experience. It uses pattern recognition, rule-based logic, and reinforcement techniques that help algorithms give preference to “good” outcomes. Quite often in healthcare, this is done through training data, in other words medical records. Machine learning can be supervised, unsupervised, semi-supervised, reinforced, self-learning and a number of other learning approaches.

  • Supervised vs. unsupervised learning: Refers to whether programs are given both inputs and outputs, this can also be described as labeled and unlabeled data. Supervised learning means that “training data” that identifies both the input (labeled) and output (answer key) so the algorithm can be “trained” to distinguish between good and bad results. Unsupervised learning, on the other hand, occurs when the algorithm is not provided with outputs (answer key) and must identify patterns, features, clusters and so forth to provide output (solutions).

  • Artificial Neural Networks: Algorithms that imitate human brains with artificially-constructed neurons and synapses. They are typically constructed in layers that each perform different functions that are then used to simulate how a brain actually works. These algorithms can obtain and process information from large quantities of data.

  • Decision Trees: A tool that maps out information according to possibilities that come from making a decision. With each decision made, there are a multitude of consequences, and a decision tree maps out the possible outcomes from making different choices in a tree-like model. AI inputs data from decision trees and determines which options will yield the best, least-costly outcomes by considering all possible options.

Applications of AI and Machine Learning in Healthcare

  • Radiology: AI assists in radiology primarily through speeding up patient diagnoses and treatment recommendations. It also can produce more accurate quantitative imaging and identify unknown characteristics that individuals with particular diseases have.

  • Imaging:  When programmed correctly, AI can identify signs of particular diseases in patient images acquired through CT scans, MRIs, and x-rays through finding abnormalities. Examples of typically identified injuries include cardiovascular abnormalities, musculoskeletal injuries like fractures, neurological diseases, thoracic complications like pneumonia, and various cancers.
  • Diagnosis: AI software has recently been able to diagnose patients more accurately than physical healthcare professionals using imaging. In the past, AI was mainly used when identifying cancers based on pictures of skin lesions. The number of AI-identifiable diseases has since expanded with technological advancements. Based on the disease diagnosed, AI tools can recommend treatment options and help develop drug treatments if they don’t already exist

  • Telehealth/Telemedicine: Telehealth/telemedicine enable healthcare to be delivered over long distances using technologies involving telecommunication and information dispersed electronically. AI uses predictive analytics to better serve rural or elderly populations from afar through diagnosing patients faster, functioning as robots to physically assist people, remotely checking in with patients to monitor progress, and reducing visits to specialty healthcare professionals.

  • Electronic Health Records: AI holds potential to simplify complicated Electronic Health Records (EHR) through making networks more flexible and intelligent by using key terms to obtain data, using predictive algorithms in EHR to warn professionals of potential diseases in patients, and simplifying data collection and entry.

  • Drug Development: Because AI can identify abnormalities in patients and what disease these physical abnormalities are linked to, it can use this information to develop drug treatments. Drug development is often slowed by human error found in the need to test several variations until it is approved by the FDA. AI speeds this process up and makes it cheaper by using pattern analysis and decision-making processes to analyze biomedical information more accurately, eliminate drug options that are likely to fail, and recruit the best patients for trials.

  • Drug Interactions: Since combining drugs is a common but potentially dangerous treatment practice, AI can warn providers about possible side effects from interactions between drugs. Penn State researchers created an algorithm using artificial neural networks that screens drug contents and look for combinations that could potentially cause harm to a patient when put into the human body. This application of AI may have large ramifications for healthcare because many patients use multiple drugs when dealing with more severe health issues and need to know that what they’re consuming won’t cause more harm.

  • Treatment Planning: Implementing AI into treatment planning has been especially heavily studied in radiotherapy. Because new technology has resulted in more options becoming available, treatment planning is more complex and labor intensive than before. AI can automate planning processes through using algorithms to identify benefits and drawbacks of treatments at much faster rates and note effects from combinations of treatments.