Important work by others has demonstrated that the interindividual variation in glycaemic responses to food are at least as variable as the differences between foods. The standard dietary advice based on glycaemic index must therefore be called into question and the importance of and individual approach to dietary advice is emphasized.
The development of a device that can monitor food intake and provide advice based on individual glycaemic responses to that food has become the focus of my research in collaboration with the Monash Faculty of IT and the Monash Institute of Medical Engineering. This work uses a smart device to obtain an image of food and then correlates this with the glycaemia response to the consumption of that food. The AI will categorize the images and predicts glycaemic AUC based on individual data. We have collected data from 36 individuals and over 3000 food item images with over 40,000 glucose/time observations. The next step is to train the AI on these data and then apply to the same 36 individuals to validate the accuracy of predictions of AUC glucose.
It is anticipated that an individual user will be able to use their smart device to image food and receive a traffic light waning of its suitability with respect to glycaemic impact prior to consumption. Such an approach may be able to improve control in subject with diabetes but may also prevent progression in those at risk of the diagnosis.
Advances in image processing will have impact in many other areas of medicine. One example is in screening for diabetic retinopathy. In a real-world study of the technology, we imaged the retinas of 236 participants. The AUC of the ROC was 0.92, the sensitivity 96.9% and the specificity 87.7%. Simultaneous grading by an ophthalmologist identified only one false negative. With the declining capacity of non- ophthalmologists to screen for retinopathy this technology is a significant advance.