RatioLogo
Back

A Digital Nutritionist for Dhaka: The OBESEYE System

In a bustling hospital in Northern Dhaka, the prescriptive power of a nutritionist is often a luxury many cannot access. For patients navigating the treacherous overlap of obesity, diabetes, and chronic kidney disease, the difference between recovery and relapse often sits on a dinner plate, yet generic dietary guidelines rarely translate to the complexities of local dietary patterns.

The Solution: A Precision Nutrition Tool

Researchers have now unveiled OBESEYE, an interpretable machine learning system designed to act as a digital nutritionist. It moves beyond simple calorie counting to provide precision nutrition—clinically validated dietary targets delivered through a mobile interface.

For a patient in a developing region, this technology could mean the difference between stable blood sugar and a metabolic crisis.

How OBESEYE Works

The system is designed to predict precise intake requirements for patients with complex comorbidities.

  • Patient Data Analysis: It analyzes the anthropometric data and clinical biomarkers of patients.
  • Study Population: The initial study was conducted with a cohort of 146 patients aged 18 to 95.
  • Core Function: It generates tailored recommendations for fluid and macronutrient intake.

Technical Performance & Model Findings

The study utilized an 80/20 Train-Test-Split, discovering that different mathematical models were optimal for predicting different nutrients.

The Optimal Models for Each Nutrient

  • Carbohydrates: A Random Forest model was the gold standard.

    • Accuracy: 86.99%
    • Root Mean Square Error (RMSE): 29.08 g

  • Protein: The LightGBM algorithm proved most effective.

    • Accuracy: 79.27%
    • RMSE: 15.95 g

  • Fluid Needs: The simplest model, Linear Regression, was superior.

    • Accuracy: 78.75%
    • RMSE: 0.39 L

Unlocking the "Black Box" with Explainable AI

Using Explainable AI (XAI) tools like SHAP and LIME, the research team interpreted how the models made their predictions.

Key Interpretability Insight

While body mass is a factor, the analysis revealed that a diagnosis of Chronic Kidney Disease (CKD) was the primary global predictor for a patient's fluid requirements.

Practical Validation & Current Limitations

The error margins for predictions were assessed against professional standards.

Clinical Tolerability & Development Hurdles

  • Tolerable Error: The error margins for fat (15.09g) and carbohydrates (29.08g) fell within "tolerable" limits defined by professional nutritionists.
  • Key Limitations to Address:
    • Small Dataset: The current dataset is relatively small (N=146), which led to some model overfitting.
    • Geographic Focus: Findings are localized to a single hospital in Dhaka and require broader validation.
    • Nutrient Scope: The system currently focuses on fluids and three macronutrients, not yet accounting for essential vitamins, minerals, or meal timing.

The Path Forward: The team views OBESEYE as a foundational step toward a scalable mHealth solution that could bridge the global gap in nutritionist availability.

Reference:
Roy, M., Das, S., & Protity, A. T. (2023). OBESEYE: Interpretable Diet Recommender for Obesity Management using Machine Learning and Explainable AI. International Journal of Recent Advances in Multidisciplinary Topics, Vol. 4, Issue 6.