Medical errors are the third leading cause of death in the United States behind cancer and cardiovascular disease. The largest proportion of medical errors involve medications.
With NLM funding, Drs. Corey Lester, Raed Al Kontar, and Xi Jessie Yang of the University of Michigan College of Pharmacy are using machine intelligence (MI) to assist in the pill verification process to help avoid dangerous and costly pharmacy dispensing errors. Their research is making sure MI is capable of conveying accurate information that encourages providers to make sound, trustworthy decisions. In doing so, they can ensure that patients get the correct pills in the correct bottle, ultimately creating safer systems that can save lives.
https://reporter.nih.gov/search/mUibh_GKPUWO-gCYSU5VRg/project-details/10434056
Transcript:
[Corey Lester, PhD] My proposal to the National Library of Medicine was to develop and design artificial intelligence, specifically a computer vision model, which could look at a picture of pills inside a medication bottle and accurately identify the contents of that bottle.
We are testing how that model can be used in collaboration with the human so that errors don't happen in the pharmacy.
[Raed Al Kontar, PhD] How do you teach a machine - if you want to a machine to differentiate between an anomaly in medical dispensing or not? What you do is - you give the machine images of anomalies, you give the machine images of good medications, and you tell them - this is good, this is bad.
When the algorithm can make this link, then if you give the algorithm a new image, the algorithm can predict - this image belongs to which class? Is it the good group or the bad group? Fundamentally, it's a matching decision done in a fancy mathematical way.
[Lester] So the way we see this being implemented is, once we have that picture of pills in the bottle, we can use this model to predict what medication it is in that image and cross-reference it with the expected or intended medication that was supposed to be filled.
In this way, we can support pharmacists in their double check verification of the patient getting the right pill in the right bottle.
[Kontar] The results were surprisingly positive and we found that our algorithm can predict with an accuracy above 99.5 percent. But what one thing that we are afraid of - even those 0.5 percent of errors can be critical.
Our hope is that, in those cases where the algorithm is wrong, the algorithm can tell us that it is not confident in its results. Then, at least the human can double check.
[Xi Jessie Yang, PhD] Our next step is to conduct the experiment with pharmacists and through that we are going to track how their trust change over time on a trial-to-trial basis.
So through the experiment we can collect participants responses, their physiological signals, and we want to use eye tracking. And by analyzing their responses and signals, we can track how their trust changes over time, how their behaviors - their performance - changes over time.
[Lester] The ultimate goal of this project is to develop safer systems that can be used to save lives and prevent medication errors from causing harm to patients.
[Yang] It means tons to me. As an engineer, I believe our ultimate mission is to propose solutions that can be used to enhance people's life. Being part of the projects doing just that.
[Lester] The National Library of Medicine's financial support of this project has really been invaluable. So really they've been instrumental in making all of this research possible.
[Kontar] Without it, we would not have had the capability to collect all this data, to do all this analysis, and generate the promising results we have right now.