Above: Image © exdez, iStockPhoto.com

One day, you’re skiing on some slopes in the more remote areas of British Columbia.   Every slope is more challenging than the last. You’re really enjoying yourself. Then suddenly, you hit a branch and take a spill.  You cannot walk… and you think your leg is broken.   

Paramedics take you to a medical facility nearby.  It's a small, remote clinic. There is only one physician (doctor) there, and she is treating another emergency. The emergency nurses are not qualified to diagnose your injury.

This scenario may sound scary. But computers and artificial intelligence may be able to one day quickly get you on the road to recovery.  

The Challenge of Diagnosing Bone Fractures

There are many types of bone fractures. For example, the bone might be cracked or broken into many pieces. Diagnosing the specific type of fracture requires medical skill. Orthopedic surgeons, doctors who specialize in the musculoskeletal system, train for years to diagnose such injuries.  

In a remote area, there likely won’t be any orthopedic surgeons. Maybe there will be an x-ray machine. If so, the staff there might be able to x-ray your injured bone.  However, there might not be any medical staff nearby who know how to analyze the image.

The x-ray of your leg is an image that can be digitized, or converted to ones and zeroes. Just as you can send an email with a picture, medical staff can send your image to an orthopedic doctor far away. However, you may have to wait hours for the result.   

Alternatively, the transmitted x-ray data can be read much more quickly by a computer.  Techniques involving artificial intelligence (AI) can analyze and categorize the data. AI is based on computer systems that can perform tasks that would normally require human intelligence.   

Artificial Intelligence for Categorizing Medical Images

To analyze images, AI systems get meaning from large amounts of photographic data.  A single high-resolution image may have one million or more pixels.  Computer programmers have trained AI systems on millions of nonmedical images. That’s trillions of pixels. Thanks to this training, these AI systems can recognize and categorize photos of things like flowers, buildings, people, and animals.

Did you know? The word “pixel” is derived from the words “picture elements.”

Physicians and medical researchers are beginning to use AI to analyze medical images, too. These medical images can help diagnoses.  

Did you know?  The first x-ray was taken in 1895, when Wilhelm Röntgen, a German professor of physics, created an x-ray image of his wife's hand.

Decision trees are a tool that can train AI systems to diagnose injuries from images. In general, a decision tree maps out the various decisions, consequences and actions of a particular issue.  It uses “leaves” to represent decisions and “branches” to represent the choice of each decision.

This simple example of a decision tree shows a decision many of us have to make a few times a day.
Image ©Ivan Lee, Let's Talk Science

AI systems can use decision trees to analyze medical images. Here’s one potential way, using a leg x-ray as an example:

First, the tree would figure out the number of pixels in the leg x-ray. Next, it would compare parts of the image to normal leg bones.  From there, thousands of decision “branches” might yield final “decision leaves.” The leaves would classify the image of the bone fracture.

As the AI systems see more and more leg bone fracture x-rays, the decision tree can improve.

Did you know? There are many other AI techniques that could analyze x-ray images. For example, neural networks are computer systems that mimic the human brain. There are also programs that mimic the process of natural selection.

Assistance for Doctors

The use of AI techniques in medicine, especially in orthopedics, is in its early days.  But techniques are advancing fast.  In one study, AI systems extracted and analyzed 256,000 images of wrists, hands, and ankles.  The AI system accurately diagnosed the type of fracture in these images 83% of the time. That’s about the same accuracy rate as an orthopedic surgeon with years of training and experience!

However, these systems cannot currently replace a doctor. A doctor would be able to do things AI wouldn’t. For example, a doctor would consider how the injury related to the patient’s overall health. A doctor could also analyze possible risks to the patient.

Someday, AI systems may be trained to make treatment recommendations based on overall health and risks, too. Until then, these systems can certainly help busy doctors analyze digital images delivered from ski resorts.

More importantly, these systems can help make sure the doctors don’t miss something.  Instead of doing the work from scratch, the doctor could double-check the system’s findings. The doctor could then evaluate your fracture remotely. Then, the two of you could discuss your injury, your general health, and various treatment options over a video conference.  

You could then return home safely, heal quickly, and begin to plan your next ski trip!

Let’s talk about it!

  • In Canada, the U.S. and much of the world, health care costs are very high. How can the AI techniques discussed in this article help with this reality?
  • Should people be required to give up their health care data to help train AI systems? Or should people be allowed to choose whether or not they want to give this information?  (Tip: Before you answer, think of some of the benefits and risks of having your personal data available to others.)
  • Should people one day be allowed to use AI tools to make their own diagnoses? Or should diagnoses always involve a doctor? Explain your answer.
  • Should pharmaceutical companies and other for-profit firms creating medical products be allowed to access people’s medical data? If so, under what circumstances?
  • If AI systems become really good at diagnosing, will we still need doctors? Why or why not? If you said yes, do you think the doctor’s job will change in any way? If so, how?

Learn More!

A Revolution in Health Care Is Coming: Welcome to Doctor You (2018)
Print Edition Leaders, The Economist.

Amazon Wants to Disrupt Healthcare in America.  In China, Tech Giants Already Have (2018)
Wee & P. Mozur, New York Times.

Artificial Intelligence (also AI) (2018)
Oxford University Press.

Artificial intelligence in healthcare: past, present and future (2017)
Jiang et al., Stroke and Vascular Neurology 2.

Artificial Intelligence for Analyzing Orthopedic Trauma Radiographs (2017)
Olczak et al., Acta Orthopaedica 88.

Stanford Medicine 2017 Health Trends Report: Harnessing the Power of Data in Health (2017)
Stanford Medicine.

The Amazing Ways How Artificial Intelligence and Machine Learning Is Used In Healthcare (2017)
Marr, Forbes.

Neural Networks and Neuroscience-Inspired Computer Vision (2014)
D.D. Cox & T. Dean, Current BIology 24.

Determining the Type of Long Bone Fractures in X-Ray Images (2013)
Al-Ayyoub & D. Al-Zghool, WSEAS Transactions on Information Science and Applications 10.

Lawrence Ostroff

I have been with SAP for several years, working in consulting, solution management, operations, and other areas. I'm currently in a go-to-market role covering strategy, planning, and execution. Before joining SAP, I worked in information technology management positions at various companies, and I hold a BA in Applied Mathematics from UC Berkeley and an MBA from Wharton. I've had the good fortune to travel around the world for business, and I've developed an interest in learning foreign languages, including Chinese and Japanese.

Je travaille pour SAP depuis plusieurs années, dans des domaines tels que la consultation, la gestion des solutions et les opérations. Actuellement, je me concentre sur la commercialisation, y compris la stratégie, la planification et l’exécution. Avant d’être embauché chez SAP, j’ai travaillé pour diverses entreprises dans le domaine de la gestion des technologies de l’information. Je suis titulaire d’un baccalauréat en mathématiques appliquées de l’Université de Californie à Berkeley et d’une maîtrise en administration des affaires de la Wharton School. Dans le cadre de ma carrière, j’ai eu la chance de voyager partout dans le monde. J’ai également développé un intérêt pour les langues étrangères, dont le chinois et le japonais.