1. Diet optimization. A machine studying mannequin for predicting blood sugar ranges after individuals eat a meal was considerably higher on the process than standard carbohydrate counting, the authors report. The algorithm’s creators used the device to compose “good” (low glycemic) and “unhealthy” (excessive glycemic) diets for 26 contributors.
“For the prediction arm, 83% of contributors had considerably larger post-prandial glycemic response when consuming the ‘unhealthy’ weight-reduction plan than the ‘good’ weight-reduction plan,” Limketkai and colleagues word. … “This expertise has since been commercialized with the Day Two cellular software on the entrance.”
2. Food picture recognition. A main problem in alerting dieters to possible dietary values and dangers going by pictures snapped on smartphones is the sheer limitlessness of attainable meals, the authors level out. An early neural-network mannequin developed at UCLA by Limketkai and colleagues achieved spectacular efficiency in coaching and validating 131 predefined meals classes from greater than 222,000 curated meals photos.
“However, in a potential evaluation of real-world meals gadgets consumed within the common inhabitants, the accuracy plummeted to 0.26 and 0.49, respectfully,” write the authors of the current paper. “Future refinement of AI for meals picture recognition would, due to this fact, profit on coaching fashions with a considerably broader range of meals gadgets that will should be tailored to particular cultures.”
3. Risk prediction. Machine studying algorithms beat out standard strategies at predicting 10-year mortality associated to heart problems in a densely layered evaluation of the National Health and Nutrition Examination Survey (NHANES) and the National Death Index.
A traditional mannequin based mostly on proportional hazards, which included age, intercourse, Black race, Hispanic ethnicity, complete ldl cholesterol, high-density lipoprotein ldl cholesterol, systolic blood strain, antihypertensive treatment, diabetes, and tobacco use “appeared to considerably overestimate danger,” Limketkai and co-authors remark. “The addition of dietary indices didn’t change mannequin efficiency, whereas the addition of 24-hour weight-reduction plan recall worsened efficiency. By distinction, the machine studying algorithms had superior efficiency than all [conventional] fashions.”