A Novel Proposal for Deep Learning-Based Diabetes Prediction: Converting Clinical Data to Image Data.

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Diabetes is a global health concern, and early detection is crucial for managing its progression.
• Traditional machine learning models face limitations with numerical medical data, hindering the application of advanced deep learning techniques like Convolutional Neural Networks (CNNs).

Purpose of the Study:

• To develop and evaluate a novel deep learning-based method for the early detection of diabetes.
• To overcome the limitations of applying CNNs to numerical medical data by converting it into an image format.

Main Methods:

• Numerical PIMA dataset was transformed into images based on feature importance.
• Three classification strategies were employed: direct CNN (ResNet18, ResNet50) on images, fusion of deep features from CNNs followed by Support Vector Machine (SVM) classification, and SVM classification on selected fused features.
• Utilized robust CNN architectures (ResNet18, ResNet50) and Support Vector Machines (SVM) for classification.

Main Results:

• The proposed method of converting numerical data into images demonstrated robustness for early diabetes diagnosis.
• Different classification strategies involving CNNs and SVMs were tested on the generated diabetes images.
• The results indicate the effectiveness of the image-based deep learning approach in enhancing early diabetes detection capabilities.

Conclusions:

• Converting numerical medical data into images is a viable strategy to leverage the power of CNNs for early diabetes detection.
• The developed image-based deep learning method shows promise for improving the accuracy and efficiency of early diabetes diagnosis.
• This approach offers a new perspective for applying deep learning to structured medical datasets in clinical practice.