You may hear the word AI and immediately picture humanoid robots from the future, like in the Jetsons and Star Wars. But what is AI exactly? As Marvin Minsky said, “artificial intelligence is the science of making machines do things that would require intelligence if done by humans.”
In fact, you may use AI and not even be aware of it. For example, wearable fitness trackers, chat boxes, auto correct & spell check, vacuum robots, and digital assistants like Alexa and Siri run on AI.
But like many, you may also have a negative view of AI. A Heartland Forward survey (which ironically used AI) finds most Americans feel negatively about AI and how it will affect their future, especially concerning jobs. However, the pollsters felt positive about AI in healthcare. About 79% said AI could have a moderate or positive impact on healthcare. This is excellent news considering using AI can help with diabetes.
AI Applications
Here is a list of AI applications in diabetes care (View full-text article in PMC).
| Applications | Description |
| Diagnosis of diabetes and it’s subtypes | Based on glycemic profile of an individual, AI can diagnose diabetes and can also classify T1DM, T2DM and GDM with physician’s level accuracy |
| Early prediction of diabetes | Diabetes onset can be predicted by various algorithms trained on electronic health record data especially predicting GDM during early pregnancy |
| Glycemic control | AI-integrated insulin pumps are utilized in a large number of studies which automate the insulin infusion rates as per the continuous glucose monitoring (CGM) |
| Glycemic events prediction | Hyperglycemia or hypoglycemia can also be predicted based on CGM data by the AI algorithms. This technique is available for commercial use as well |
| Novel markers identification | AI models identify various predictors for diabetes such as age, waist, BMI, hypertension, etc. |
| Diabetes retinopathy prediction | Risks for diabetic retinopathy can also be predicted using clinical dataset, thus help initiating early management |
| Diabetes complications diagnosis | Serious diabetes complications such as diabetic foot, diabetic retinopathy and patients at risk of rehospitalization can also be identified by various AI/ML approaches used in various studies |
Below are in-depth examples of how AI helps diabetes.
Detecting Diabetes
One way AI helps with diabetes is by detecting it. A study published in the Mayo Clinic Proceedings: Digital Health medical journal used voice analysis as a prescreening or monitoring tool for type 2 diabetes by examining the differences in voice recordings between nondiabetics and type 2 diabetics.
Up to 200,000 distinct characteristics can be present in a person’s voice, which AI algorithms can filter through to identify particular vocal patterns that match specific symptoms, such as diabetes.
A smartphone application detected significant differences in vocal pitch and intensity between nondiabetics and type 2 diabetics. While the trial theorized that the voice analysis could serve as a prescreening tool, researchers noted further research is needed and that it’s not a replacement for medical diagnosis (https://www.dw.com/en/how-ai-can-detect-diabetes-with-a-10-second-voice-sample/a-67400425#:~:text=Artificial%20intelligence%20can%20analyze%20speech,comes%20with%20a%20warning%20label.&text=Medical%20diagnostic%20tools%20using%20advanced%20voice%20analysis%20are%20becoming%20increasingly%20precise.).
This is excellent news, considering 240 million adults have undiagnosed diabetes worldwide. In the future, there will also be a dramatic increase in diabetes, thus making early detection crucial.
Monitoring Diabetes
According to a report from Teladoc Health, a leader in virtual care, patients who receive personalized health “nudges,” which are mobile notifications, reduced their A1C levels from 8.2% to 7.8%. Also, patients receiving weekly emails with personalized next-best actions are 50% more likely to work with a health coach. Thus, the research shows a positive correlation between program engagement and improved clinical outcomes (https://www.aha.org/aha-center-health-innovation-market-scan/2024-07-02-type-2-diabetes-patients-can-benefit-ai-powered-nudges-report).
Detecting Diabetic Retinopathy
“AI isn’t just a tool; it’s a partner in care. For example, AI can detect early signs of eye damage from diabetes in retinal images as accurately as specialists, which is critical for preventing blindness,” states Dr. Ling Gao from the Central Laboratory at Shandong Provincial Hospital. This happened in Australia when research showed that an accurate AI-powered camera could prevent up to 40,000 cases of blindness from diabetic retinopathy while reducing healthcare costs by being cheaper than traditional tests and not needing specialized training.
The AI camera uses thousands of images of healthy and diabetic retinopathy eyes to detect signs of the disease. It can also identify undiagnosed people with diabetes (https://www.insightnews.com.au/ai-scan-for-diabetes-has-potential-to-save-sight-and-money/).
FDA-Approved AI Tools
The FDA has approved AI tools for early intervention and management of diabetes. For example, in 2018, the FDA approved Digital Diagnostics, which has a high diagnostic accuracy in detecting diabetes retinopathy in retinal screening images. The same year, the FDA approved the Guardian Connect System to interpret biomedical data and predict a hypoglycemic attack an hour in advance (https://www.nature.com/articles/s41746-024-01034-7).
Continuous Glucose Monitor (CGM)
A continuous glucose monitor uses AI algorithms and machine learning (ML). Machine learning identifies patterns in data and uses the data to take action or solve a problem. Here, it decides how much insulin to administer based on input from a glucose sensor that measures blood sugar levels.
Two leaders in diabetes technology, Roche and Dexcom, have new and exciting AI-based CGMs.
Roche’s Accu-Chek is launching the SmartGuide wearable sensor and phone app to predict high and low blood sugar. Roche’s real-time CGM will offer blood sugar readings every 5 minutes and AI-based glucose predictions from 30 minutes to 2 hours in the future. It will also inform the wearer if there is a risk of hypoglycemia overnight.
Dexcom is releasing a Generative AI (GenAI) platform named Stelo, the first over-the-counter glucose biosensor to use GenAI to offer actionable weekly insights. Due to its 93% accuracy rate, Stelo does not require a fingerstick blood reading. The Stelo app on a smartphone shows glucose levels and target range to reveal patterns in glucose levels. This enables wearers to make informed choices. Also, personalized insights on the app suggest next steps for glucose management.
Closed Loop System or Artificial Pancreas
Many Type I diabetics, such as myself, use a closed-loop system, which consists of a CGM and insulin pump that communicate with each other. It’s called an artificial pancreas because the system mimics the function of a pancreas by monitoring blood sugar levels on the CGM and then using an algorithm to deliver insulin via the pump based on the blood sugar readings.
Artificial pancreas picture from MD+DI Qmed+
Conclusion
Not all AI is negative or to be feared. It’s currently being used in diabetes management to detect diabetes, monitor blood sugar levels, prevent eye disease, personalize recommendations, and predict low and high blood sugar levels. As Dr. Ling Gao states, “While AI won’t replace human clinicians, it empowers them to make faster, smarter decisions—ultimately transforming diabetes from a one-size-fits-all disease into a condition managed with precision and foresight.”
Quote
