In 2017, IBM’s AI research team introduced a visual recognition food model which can identify food based on a picture. Taking it a step further, IBM was recently granted a patent for estimating blood-sugar levels using images of meals. Aside from just simply identifying the food, deep learning was used to analyze the meal image, time-period, and user’s previous blood-sugar levels to provide future blood-sugar levels.
According to the 2020 National Diabetes Statistics Report, about 34.2 million people or 10.5% of the US population have diabetes with at least 77.8% receiving diabetes care. In IBM’s granted patent, which particularly relates to the field of computing and medical technology, they claim a faster way to identify sequential glucose levels from meal images. This technique may help prevent rapid fluctuations in a diabetic’s glucose levels that can be fatal. The invention aims to reduce the burden on diabetics by reducing the frequency of blood extraction for glucose level testing. In addition, because of the deep learning approach, a more personalized diabetes management may be implemented.
In an embodiment mentioned in the patent, glucose level determination may include 2 phases, a training and a prediction phase. The training phase uses inputs such as previous meal photos, time and data from a user’s continuous glucose monitor. By implementing a deep learning network, the training phase outputs an estimated meal factor that is used by the prediction phase. This prediction phase also involves deep learning which then predicts the sequential blood glucose level. If, for example, meal photos were the only data available, the system can automatically identify that and make necessary adjustments to be able to predict the user’s sequential glucose levels. The program can be run on networked platforms such as a personal computer, mobile device, or an automobile, as long as it is connected to the cloud.
In a 2019 press release from Medtronic, a similar feature of predicting hypoglycemia in Medtronic’s personal diabetes assistant app was launched in partnership with IBM Watson Health. The feature was described as a “weather forecast for people with diabetes so they can better prepare for their day” by predicting an expected low glucose level event and notifying a user through the app.
This prediction tool can further be used to understand glucose activity in the body and be an instrument to identify future diseases.
The featured top patent, “Estimating sequential blood-sugar levels using images of meals”, published as U.S. Patent No. 11,030,493, was filed on September 20, 2018 to the U.S. Patent and Trademark Office with application number 16/137,144. The inventors of the granted patent are Takuya Goto, Yutaka Oishi, Chiaki Oishi, Shuji Umehara, and Masaki Saitoh. The patent is currently assigned to International Business Machines Corporation.