MIT Develop Sensor an AI to Detect Errors in Self-Administered Medication
29-03-2021 | By Robin Mitchell
Recently, researchers from MIT have combined a radio system with AI to look for errors in self-administered medicines by patients. What problems do self-administering medicines have, how does the MIT-developed system work, and how can it be potentially expanded in the future?
What problems do self-administered medicines face?
For the majority of the population, illnesses that require medication either require a visit to a GP for diagnosis and treatment, or a visit to a hospital for longer treatment cycles. However, many live with chronic and deadly conditions, but far too impractical to be treated by doctors daily.
Two examples of such conditions would include diabetes and asthma. Diabetes (depending on the type), is a condition whereby the body either struggles to produce insulin or has problems regulating it correctly. Either way, if not properly maintained, the result of non-treated diabetes includes loss of limbs, organ failure, and death.
Fortunately, diabetes can be treated with the use of self-administered insulin injections, but these injections must be administered correctly. The first factor that must be considered is the correct timing and dosage of insulin; insulin is often taken after meals, but patients should still monitor their sugar levels. The second factor that must be considered is air bubbles in insulin pens. Using an insulin pen must be done in a specific manner to ensure that there is no trapped air in the syringe.
If insulin is administered incorrectly, patients can suffer from serious complications including hyperglycaemia, hypoglycaemia, and gas embolisms which can cause stroke or death. It can be seen that self-administered medicines for chronic conditions can carry serious risk, and therefore patients must take extreme precautions when administering medication.
MIT Researchers Combine Radio Waves with AI
Recently, MIT researchers announced a system that combines radio wave modulation with AI to determine if patients are correctly administering medication. The system is made up of two parts; a radio transceiver station similar to a Wi-Fi router and an AI neural network.
The radio system emits modulated radio waves with a specific pattern, and when patients are within 10 meters of the system the reflected radio waves (from the patient) are monitored. The system then passes the detected radio waves into an AI neural net which is trained with various activities including correct usage of medicine, eating, and walking.
If the user administers medication, the AI can detect this activity, and since self-administered medication follows a very specific pattern (movement and timing), the AI can determine if the medicine is being administered correctly. If the medicine is not being administered correctly, the system can alert medical professionals of the incident. For example, the system developed by MIT can not only recognise the use of an insulin pen, but determine for how long the pen is used for. Thanks to the use of AI, the system has been repeatedly trained to provide a success rate of 96 percent for insulin pens and 99 percent for inhalers.
How AI Can Help Solve Most Problems
The system developed by MIT demonstrates a recurring pattern of engineers turning to AI to solve extremely difficult problems. One of the biggest advantages of AI is that it is able to infer results from the most seemingly obscure data. As such, AI is being incorporated into more than just image and language interfaces; it’s also being used in sensors.
Traditional sensors produce results that are extremely well defined and obvious. For example, a light sensor typically has a linear relationship between the light falling on it and the voltage (or current) it produces. However, researchers are creating modern sensors that produce outputs that often appear erratic and unclear, but AI systems can show a correlation between the output and the desired stimuli.
Fundamentally, what AI does to a problem is to move it away from code-defined behaviour to learned behaviour. A traditional system requires complex code to describe how to infer results, while an AI system needs large amounts of data and result to be able to determine how to come to a correct conclusion.
Read More
