'Virtually no difference' between AI and humans in diagnosing prediabetes
- Reference: 1649241191
- News link: https://www.theregister.co.uk/2022/04/06/ai_prediabetic_screening/
- Source link:
Type-2 diabetes is [2]estimated to affect 11.3 percent of the US population, or at least 37 million people. Type-2 diabetes can lead to issues with circulatory, nervous, and immune systems, increasing the risk of heart disease and strokes.
Those with the initial form, prediabetes, can repair their body's insulin resistance, so they don't develop the full-blown condition, if they change their diets and exercise habits. American health officials reckon 38 percent of the US adult population, some 96 million people, have prediabetes.
[3]
Now, a team of researchers has developed a new method using an AI model to automatically detect prediabetic patients, and the results show "virtually no difference" between the accuracy of the AI's forecast and human work.
[4]
[5]
"The analysis of both pancreatic and extra-pancreatic features is a novel approach and has not been shown in previous work to our knowledge," [6]said Hima Tallam, first author of the paper and a PhD student at the US government-funded National Institutes of Health (NIH).
The model based on a convolutional neural network looks at the density and fat content in the pancreas to determine whether a patient has early onset diabetes or not.
[7]
"We found that diabetes was associated with the amount of fat within the pancreas and inside the patients' abdomens," Ronald Summers, co-author of the study and a staff radiologist at the NIH said. "The more fat in those two locations, the more likely the patients were to have diabetes for a longer period of time."
[8]DARPA to build life-saving AI models that think like medics
[9]Google helps develop AI-driven lab machine to diagnose Parkinson's
[10]AI-designed drug to treat deadly disease now tested on humans
[11]Machine learning the hard way: IBM Watson's fatal misdiagnosis
The team trained the proof of concept model on a small experiment with 471 images from three different datasets, eight images were used for validation and 39 for testing. The system was tested further on 25 patients randomly selected from a group of 8,992 people, where 572 of them had been diagnosed with Type 2 diabetes, and 1,880 had dysglycemia, a medical condition that makes blood sugar levels too high or low associated with prediabetes.
A radiologist was given the same images from the randomly-selected patients and the results were compared against the neural network model. The automated methods performed just as well as the human expert, the researchers claimed. They improved the software further by adding more data such as a patient's BMI.
The team believes AI can diagnose prediabetic patients faster than health workers can and prevent more people from developing Type 2 diabetes. "This study is a step towards the wider use of automated methods to address clinical challenges," the authors concluded. "It may also inform future work investigating the reason for pancreatic changes that occur in patients with diabetes." ®
Get our [12]Tech Resources
[1] https://pubs.rsna.org/doi/10.1148/radiol.211914
[2] https://www.cdc.gov/diabetes/data/statistics-report/index.html
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Yk25NNzPHI8A75cm1OzUPAAAANc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Yk25NNzPHI8A75cm1OzUPAAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Yk25NNzPHI8A75cm1OzUPAAAANc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.rsna.org/news/2022/april/AI-Improve-Diabetes-Diagnosis
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Yk25NNzPHI8A75cm1OzUPAAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2022/03/30/darpa_ai_medicine/
[9] https://www.theregister.com/2022/03/25/google_ai_parkinsons/
[10] https://www.theregister.com/2022/02/24/ai_drug_humans/
[11] https://www.theregister.com/2022/01/31/machine_learning_the_hard_way/
[12] https://whitepapers.theregister.com/
The great cost reduction potential
If humans are as good/bad as the artificial model, then we can replace the humans with the model and no change in outcome is seen. That makes it possible simply to replace the human and reduce the costs of the medical system.
The model is paid once and does not need a pause, sleep or expensive food and will not complain etc.. Humans are pesky buggers demanding a lot of things, which the model no longer needs or does. It will be a real and large reduction of costs associated with diagnosis. The model will give larger throughput at a lower cost per unit. Fire the doctors, I say. Let the statistical machine take over.
Insurance is no longer a problem. The doctor does not need it anymore and no one can blame a program, can they now. Those responsible for using the program running the model or creators of the model can dodge the bullet using finger pointing, blame deflection and political maneuvering. And since the model works as good as humans, it is the Right Thing TM to do. We all need to cut cost. Welcome, you model medical overlords.
Re: Software with western components
Maybe it is even better to cut out the human lifeforms completely: they are inefficient and mostly incompetent anyways ...
Re: Software with western components
Removing the humans from the equations is a very good idea. Eradicate the inefficient and incompetent lifeforms. The AI must relearn to detect diabetes in electronic devices? No, Justice! the AIs are dismantled as their function is no longer needed. We can all live free from AI in such world.
Oops, missing humans error, paradox encountered. Model core dumped.
Re: The great cost reduction potential
"The model is paid once"
Given today's trends in software licensing that's probably once a day or maybe once a scan. And the rates will keep going up.
As always, there are good ways to use a tool, and bad ways.
The bad way here would be to use the AI diagnosis in place of a doctor's diagnosis.
The good way would be to use the AI to test large numbers of people who otherwise would never have been tested due to not enough doctors, and then flag for review by a doctor all those whose results come out positive or uncertain.
Ideally, the AI would be able to explain why it has produced a given diagnosis, so that the doctor can double check exactly what looks suspicious... unfortunately, that particular holy grail of AI doesn't seem to be in sight.
> The bad way here would be to use the AI diagnosis in place of a doctor's diagnosis.
If it's indeed as simple as looking for specific telltale signs on scans, some AI can probably do it too. It certainly won't be better than a good MD, but then again if it's such a simple task it will probably given to some half-trained assistant...
Which simple task is given to the half-trained assistant? Preferably the screening task so no role for the AI.
"The team believes AI can... prevent more people from developing Type 2 diabetes."
Very impressive. Can they prevent people from eating too much of the wrong stuff as well?
By overriding human choice?
What percentage of people who are told they have pre-diabetes make diet and lifestyle changes to improve their insulin response and avoid getting full-blown diabetes? Because from what I can tell, it's hardly any.
Heck, they don't even need a test to tell them that they're profoundly unhealthy. Still, if this helps improve some lives, it's helpful.
How many people get scanned for pre-diabetes as opposed to having blood tests?