DiabetesXpertNet: An innovative attention-based CNN for accurate type 2 diabetes prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Type 2 diabetes mellitus (T2DM) is a growing global health concern with severe complications.
• Traditional prediction models face challenges like data imbalance, high dimensionality, missing values, and outliers, limiting their effectiveness.
• Accurate early prediction of T2DM is crucial for timely intervention and management.

Purpose of the Study:

• To introduce DiabetesXpertNet, a deep learning framework specifically designed for enhanced prediction of Type 2 diabetes mellitus using tabular medical data.
• To address limitations of existing models by incorporating specialized attention mechanisms and feature enhancement techniques.
• To improve the accuracy, interpretability, and fairness of T2DM prediction models.

Main Methods:

• Developed DiabetesXpertNet, a convolutional neural network (CNN) framework with dynamic channel attention for prioritizing key clinical features.
• Integrated a context-aware feature enhancer to capture sequential relationships in structured medical data.
• Employed advanced preprocessing: mean imputation, median outlier replacement, mutual information/LASSO for feature selection, and logistic regression-based class weighting to handle data challenges.

Main Results:

• DiabetesXpertNet achieved high performance on PID and Frankfurt Hospital datasets: 89.98% accuracy, 91.95% AUC, 89.08% precision, 88.11% recall, and 88.01% F1-score.
• Demonstrated significant improvements over traditional machine learning models in precision (+5.1%), recall (+4.8%), F1-score (+5.1%), accuracy (+6.0%), and AUC (+4.5%).
• Outperformed other CNN models with notable gains in precision (+2.2%), recall (+1.1%), F1-score (+1.2%), accuracy (+1.9%), and AUC (+0.6%).

Conclusions:

• DiabetesXpertNet proves robust and interpretable, outperforming existing methods for Type 2 diabetes mellitus prediction.
• The framework's specialized architecture and preprocessing techniques effectively handle challenges in tabular medical data.
• DiabetesXpertNet represents a promising advancement for early and accurate clinical diagnosis of Type 2 diabetes.