How Does AI Diagnose Images?
By Kyle Henson  |  Apr 15, 2022
How Does AI Diagnose Images?
Image courtesy of and under license from Shutterstock.com
Based on how humans can tell the similarities and differences between certain objects, mathematical features and classifiers can be used in algorithms so that AI can come to the same types of conclusions when analyzing more complex problems. But what is the best way to use features and classifiers to minimize the risk of an algorithm returning false positives or false negatives?

DALLAS, TEXAS - This is a question which has perplexed me ever since AI became a hot topic in the medical imaging space. I have met with countless radiologists, experts, and data scientists in an effort to answer this one seemingly simple question. The answers I received range from the simple (trial and error) to the immensely complex rambling of someone who didn’t know the answer but wanted to sound really smart and confuse me in the process. 

In the end, the way AI analyzes an image is in fact not all that different from a human. AI uses features and classifiers to determine the mathematical probability of a pathology.

Let’s set aside the complex terminology for a moment and ponder a simpler question. How do you tell the difference between a baseball, a golf ball, a basketball, and a football? Let’s start by looking at their shapes. Three of these things are round, and one is oblong. Next, let’s add color: two are white, one is brown, and one is orange. If we added curving cross-sectional lines, we would conclude that two of the balls had them and two did not. What about dimples or laces? Only one ball has dimples, while one has laces.

Now, what if I told you that we can mathematically define each of those descriptions (shape, color, wavy lines, dimples, and laces)? The roundness of an object can be defined in a computer and then measured on a scale from 0-1, with zero being not round at all (like a straight line), and 1 being a perfect circle. Color can also be defined in a similar way, with white being 255,255,255 in the RGB standard. This enables us to measure the difference between white and whatever color is being seen. For example, using the Windows standard orange, you could define orange as being 255,192,000 . Again, you can now rate the color of an object on a scale fr

The content herein is subject to copyright by The Yuan. All rights reserved. The content of the services is owned or licensed to The Yuan. The copying or storing of any content for anything other than personal use is expressly prohibited without prior written permission from The Yuan, or the copyright holder identified in the copyright notice contained in the content.
Continue reading
Sign up now to read this story for free.
- or -
Continue with Linkedin Continue with Google
Comments
Share your thoughts.
The Yuan wants to hear your voice. We welcome your on-topic commentary, critique, and expertise. All comments are moderated for civility.