Medically reviewed by Susan Kerrigan, MD and Marianne Madsen
Radiology is the study and use of high-energy substances and emissions. It has become an inextricable part of numerous fields of medicine. Using radiological tools such as an X-ray, MRI, or CT scan, doctors can now see the internal structures of the body with infinitely greater clarity, an advantage that translates directly into more efficient and effective treatment efforts.
Machine learning is another tool that has entered the medical community relatively recently but is already showing significant promise. This growing area of medicine involves programming software to recognize certain specific patterns based on a massive database of previous cases. The hope is that by immediately comparing thousands of similar pictures, the machine learning engine will be able to spot the relevant signs and symptoms of particular diseases more reliably than a human.
How does it work?
Artificial intelligence in radiology works by “teaching” the software to be able to “look” at a particular image and recognize the problem. For example, the software could examine an x-ray and “see” a broken bone, even one that might not be seen by a human. It can do this by scanning through its database of images of broken bones and “recognizing” another image of a broken bone when it “sees” one. The programs that recognize these patterns are called algorithms; the more of something they see, the more likely it is that they will be able to spot another instance of the same problem.
Benefits
Humans cannot always process the considerable amount of information contained in or associated with an X-ray, MRI, or CT scan. By using a computer to analyze the image too, doctors can ensure that all of this data is being properly compared to existing databases, leaving them free to focus their efforts on the few images that the software may flag as problematic. Machine learning databases can be quickly and constantly updated; getting new information to a human analyst might take a significant amount of time.
Additionally, the use of computers in radiology brings all the benefits of automation in other sectors–a machine can do an important task repeatedly and reliably for nearly indefinite amounts of time.
Risks
Machine learning is in the developmental stage in many of its more advanced applications; although it is steadily being refined and improved, complicated tasks like radiological analysis are not yet totally reliable. There is also the issue that for most machine learning, the software will only flag as an anomaly one specific condition; if the patient has a problem that the software has not been set to look for, it will arouse no suspicion from the algorithms checking the photos.
Conclusion
Machine learning shows significant promise as an aid to radiologists everywhere, promising faster and more reliable alerts to any problems. It is important to remember that AI is an aid to finding problems and not a replacement for a doctor.
Written by Shlomo Witty
