AI-DRIVEN HEALTH RESEARCH
Can Artificial Intelligence Predict Glycaemic Responses to Food from Food Photographs Using a Mobile App?
This study explored the feasibility of developing AI models to predict individual glycaemic responses to food from mobile app photographs. Leveraging continuous glucose monitoring, it highlights the potential for personalized diabetes management and the need for high-quality datasets in future research.
Executive Impact: Key Metrics & Findings
Insights from this research demonstrate the potential of AI in personalized health, showcasing significant data collection, usability, and model performance milestones.
0
Food Images Collected
0
Glucose Readings
0
App Usability (SUS Score)
0
Highest Model Accuracy
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Study Design and AI Pipeline
This study aimed to develop AI models to predict individual glycaemic responses to food. It utilized continuous glucose monitoring (CGM) devices and a custom mobile app to collect meal photos. Data preprocessing involved aggregating glucose readings and food intake data across 30, 45, and 90-minute intervals, applying statistical measures such as peak glucose, AUC, velocity, and deviation. Food items were then labelled as 'Green' (healthy) or 'Not Green' (poor) based on their glycaemic impact. A pre-trained MobileNetV2 model was fine-tuned for classification, and model performance was evaluated using accuracy, recall, precision, and F1-score across different participant groups.
AI Model Development Pipeline
Data Collection
→
Data Preprocessing
→
Feature Engineering
→
Data Labelling (Healthy/NotHealthy)
→
Model Training (MobileNetV2)
→
Model Evaluation
Performance and Usability Insights
The models generally showed improved performance with longer post-meal intervals (90 minutes), with the peak glucose feature proving most effective. The mobile app received a positive System Usability Scale (SUS) score, indicating good user experience despite challenges such as remembering to log meals and initial food detection inaccuracies. Personalised models showed strong performance for certain participants with sufficient and balanced data, highlighting the critical need for larger, high-quality datasets and longer-term studies to enhance precision and recall for all users.
80
System Usability Scale (SUS) Mean Score - Indicating 'Good' Usability for the Mobile App
| Model & Population | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| MobileNetV2+Peak Glucose (All Participants) | 0.86 | 0.50 | 0.50 | 0.49 |
| MobileNetV2+Peak Glucose (Diabetic Population) | 0.95 | 0.48 | 0.49 | 0.49 |
| MobileNetV2+Peak Glucose (Healthy Population) | 0.75 | 0.48 | 0.49 | 0.47 |
| MobileNetV2+Peak Glucose (Personalised P035) | 0.94 | 0.75 | 0.97 | 0.82 |
Estimate Your AI Transformation ROI
Quantify the potential efficiency gains and cost savings from implementing AI solutions inspired by this research in your enterprise.
Advanced Efficiency Calculator
Your AI Implementation Roadmap
A structured approach to integrating advanced AI into your operations, drawing lessons from pioneering research.
Phase 1: Discovery & Strategy Alignment
Goal: Define clear AI objectives and assess current infrastructure. This phase involves detailed stakeholder interviews, identification of high-impact areas, and setting measurable success metrics for personalized health AI, similar to the initial data collection and problem definition in the research.
Phase 2: Pilot Program & Data Integration
Goal: Implement a small-scale AI solution and establish robust data pipelines. Mirroring the study's data collection and preprocessing, this involves setting up continuous data feeds (e.g., from wearables, food logs), ensuring data quality, and training initial models with representative datasets.
Phase 3: Iterative Development & Scaling
Goal: Refine AI models and expand deployment based on feedback. Emulating the model evaluation and refinement cycle, this phase focuses on improving model accuracy, handling class imbalances, and optimizing for real-world performance as the solution scales across the enterprise.
Phase 4: Ongoing Optimization & Impact Measurement
Goal: Continuously monitor, optimize, and measure the long-term impact of AI. This involves A/B testing, integrating user feedback (like app usability studies), and quantifying ROI through improved health outcomes, operational efficiencies, and user satisfaction, reflecting the long-term study needs identified.
Ready to Transform Your Enterprise with AI?
Leverage insights from cutting-edge research to develop personalized AI solutions that drive real-world impact. Our experts are ready to guide you.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat