Design of a Low-Complexity Deep Learning Model for Diagnosis of Type 2 Diabetes

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Diabetes mellitus is a growing global health concern.
• Diabetes complications include cardiovascular disease and neurological damage.
• Early diagnosis is crucial for effective diabetes management.

Purpose of the Study:

• To develop a low-complexity Deep Learning (DL) model for type 2 diabetes diagnosis.
• To evaluate the performance of various DL architectures for diabetes detection.
• To identify an optimal DL model balancing accuracy and computational complexity.

Main Methods:

• Utilized the PIMA Indian Diabetes Dataset (PIDD) for model training and validation.
• Designed and compared multiple DL models, including Convolutional Neural Networks (CNN) and Multi-Layer Perceptrons (MLP).
• Conducted an accuracy-versus-complexity analysis to select the most efficient DL architecture.

Main Results:

• The proposed hybrid CNN+MLP model demonstrated superior performance.
• Achieved a high diagnostic accuracy of 93.89% for type 2 diabetes.
• Outperformed other evaluated DL models in terms of accuracy and efficiency.

Conclusions:

• The developed hybrid DL model offers a promising solution for diabetes diagnosis.
• Its low complexity and high accuracy make it ideal for integration into wearable devices.
• Enables potential for real-time health monitoring in Internet of Things (IoT) applications.