Current Uses of AI in Healthcare - Cover

Current Uses of AI in Healthcare

Health information technology is an exciting industry for the potential application of AI. Practical applications are not widely used or available, but early adopters have made some interesting first steps. So let’s cut through speculation and theoretical applications and look at some verified uses in the industry.

Mayo Clinic

Mayo Clinic researchers have created an AI-enabled electrocardiogram. This imaging device is designed to detect atrial fibrillation (AF). AF is hard to diagnose, especially when the heart presents a normal rhythm during a test. The convolutional neural network system reviewing electrocardiographic image data in this study correctly identified AF patterns 90% of the time.

Memorial Sloan Kettering

IBM, in collaboration with Memorial Sloan Kettering Cancer Center, has worked on a cognitive computing system to extract data from documents via natural language processing (NLP). This NLP process was applied to large, unstructured data sets of oncology patients. The recommendations of IBM Watson were then compared to the recommendations of a human multidisciplinary tumor board. The AI system and the human physician board concurred on treatment plans 73% of the time in the study. The conclusion of the study was that the AI system required further refinement of more complex cases to improve treatment recommendations.


Engineers at Stanford have developed a deep convolutional neural network trained to review and categorize skin lesions from image data. When put to the test with professional dermatologists, the AI was on par with medical professionals in identifying skin cancer subtypes from given image sets. The practical use envisioned by the authors of the study was to develop an at-home app that everyday users could leverage to photograph their own skin lesions and determine whether they should see a dermatologist about a potential case of skin cancer.

Battelle Memorial Institute

Injuries of many types can result in paralysis when signal pathways between the brain and muscles become disrupted. Medical researchers are working on devices that can aid in rerouting these electrical signals, thereby circumventing the damage done to the nervous system and allowing persons with paralysis to regain motor control. However, for the brain to be able to send a signal that these neuroprosthetic devices can interpret, the team in this study needed to examine motor cortex activity from a quadriplegic person and decode these signals into intended movements.

This neuronal-activity data was decoded by machine-learning algorithms. Armed with this data, the team was able to build and implement a high-resolution neuromuscular electrical stimulation system. The system allowed the participant to make isolated finger movements, and he had continuous cortical control of six different wrist and hand motions. The participant was able to use the system to complete functional tasks related to daily living, such as grasping, manipulating, and releasing objects.

More Insights

Other instances of AI being used in healthcare IT can be explored at length in a recent Stroke and Vascular Neurology report. I found a few key insights on the state of AI based on trends found in this report and on data gathered by KLAS:

  1. These technologies are not yet intended to replace human involvement. Instead, they are meant to supplement human capacity and empower physicians to perform their work at higher levels of confidence and at greater speed.
  2. These technologies are being developed mostly for cardiology, neurology, and oncology specialties. This may be because these specialties rely on the detailed interpretation of diagnostic-quality images. The capacity for computers to store, recall, and process vast quantities of data is one prime reason for AI to be connected to image interpretation. One key step forward in the work will be to instruct machines on how to correctly interpret imaging data. Future products of interest will need to integrate with PACS solutions and other imaging hardware in order to be of clinical use and to make financial sense.
  3. These technologies are still in the early-adoption stage. The tasks already performed verify the utility and capacity for these innovations, and now the industry is working on formalizing individual innovations into early products and services. This trend will likely continue for a few years at least before these technologies are seen in an early-majority phase.

At KLAS, we are fascinated to see where the interaction of AI and healthcare will go.