Artificial Intelligence and Medical Imaging
Artificial Intelligence (AI) seems to be a popular media topic these days, and with good reason. The past decade has seen a phenomenal growth in deployed AI applications, many of which are having or will soon be having profound effects on society. Three are three things driving the rapid advances in the field:
- Fast, inexpensive, and plentiful computers (i.e., speed);
- Lots of data (i.e., storage/memory); and
- Data analysis (i.e., learning).
This is the technology triumvirate of the modern information age. It’s a powerful combination, and one that is fueling the rapid progress in AI research and applications. Let’s look at each one briefly:
- The need for speed… Many AI computations have enormous computational requirements. Much of this comes from the need to converge to an answer – a program that runs for a day might compute a good answer, but one running for 10 days might lead to a better answer. Technology advances have enabled key applications to run 1,000,000 times faster than they could a decade ago!
- Size matters… at least when it comes to data. Large data sets are important not because they illustrate the common scenarios, but because they are critical for finding the uncommon (rare) scenarios. For example, a doctor may see thousands of children that have a headache, but rarely is the cause a brain tumor. You need lots of cases to be able to see and identify the exceptions.
- Fool me twice, shame on me… Fast computers can analyze large data sets and learn from them. There are many machine learning models, but most of them are adaptive in that each new data point allows the model to refine itself. In other words, the AI learns from its mistakes, and from the mistakes made by others.
The combination of the three has allowed some AI applications to become super-human. For example, computers can now out-perform humans at recognizing and identifying people’s faces. And what is a face to a computer? Just a combination of learned patterns.
Computers are on the cusp of creating a revolution in medical imaging. What is a tumor on a CT scan? To a computer it is just a combination of learned patterns, a different kind of “face”, if you will. Given an appropriate data set and one of sufficient breadth (size), computers can be “taught” to recognize medical conditions from an image. Geoffrey Hinton, the 2019 Turing Award winner (the Nobel Prize of computing science) said in 2017 that:
“We could build in a system that would take every missed diagnosis—a patient who developed lung cancer eventually—and feed it back to the machine. We could ask, What did you miss here? Could you refine the diagnosis? There’s no such system for a human radiologist.”
A recent paper in the journal of European Experimental Radiology (December, 2018) was more blunt:
“AI will surely impact radiology, and more quickly than other medical fields. It will change radiology practice more than anything since Roentgen.”
These are still early days for AI applied to medical imaging, but the early results are promising, and the research effort being invested in this area is growing in leaps and bounds. Does this mean imaging experts will soon be out of a job? No. Think of the AI as a second opinion. Leave the final decision to the human, but use the AI as a tool to aid the human in coming to the best decision. Man and machine need to work symbiotically.
To prepare for the AI revolution in your field, here is advice given in the recent journal paper:
“An updated radiologist should be aware of the basic principles of [AI] systems, of the characteristic of datasets to train them, and their limitations. Radiologists do not need to know the deepest details of these systems, but they must learn the technical vocabulary used by data scientists to efficiently communicate with them. The time to work for and with AI in radiology is now.”
Jonathan is a professor in the Department of Computing Science at the University of Alberta. His research is in the area of artificial intelligence, and he is well known for using games to demonstrate his work.