Integration of Clinical Criteria into the Training of Deep Models: Application to Glucose Prediction for Diabetic People

Abstract

Traditional neural network training methods in healthcare often overlook clinical requirements, resulting in models that may achieve high accuracy but lack clinical relevance. This study tackles that limitation in the context of glucose prediction for individuals with diabetes. The authors introduce a novel training methodology that balances statistical accuracy with clinical constraints established by health authorities. By progressively shifting focus from pure prediction accuracy to clinical acceptability, the proposed approach employs a custom loss function designed specifically for glucose forecasting. Evaluations on datasets from both type-1 and type-2 diabetes patients demonstrate that the method significantly improves clinical validity. Notably, the framework identifies models that not only maintain high predictive accuracy but also satisfy crucial clinical standards.

Pegah Khorasani

Graduate Research Associate

As a doctoral candidate at Arizona State University (ASU), I am currently conducting research under the guidance of Professor Hassan Ghasemzadeh at the Embedded Machine Intelligence Lab (EMIL). My research interests encompass a range of topics, including machine learning, health monitoring system development, and mobile health.