Transfer Learning-driven Biofield Image Analysis for Predictive Modeling of Diabetes.

Area of Science:

• Biofield imaging analysis
• Computational diagnostics
• Medical imaging and machine learning

Background:

• The human biofield reflects physical and emotional health.
• Biofield therapies (e.g., Reiki) use this information for assessment.
• Gas Discharge Visualization (GDV) and Polycontrast Interference Photography (PIP) are key biofield imaging techniques.

Purpose of the Study:

• To detect pancreatic energy imbalances via biofield imaging.
• To classify subjects as diabetic or nondiabetic based on biofield patterns.
• To evaluate biofield information's potential in energy-based diagnostics.

Main Methods:

• Applied color-based clustering for image segmentation.
• Developed an ensemble deep learning framework using ConvNeXtBase and ResNet50.
• Utilized grid search for hyperparameter optimization and 5-fold cross-validation.

Main Results:

• The ConvNeXtBase + ResNet50 ensemble achieved 99.12% accuracy.
• Individual models showed high accuracy (ConvNeXtBase: 97.93%, ResNet50: 96.28%).
• Area Under the Curve (AUC) values exceeded 0.99, confirming strong reliability.

Conclusions:

• Convolutional Neural Network (CNN) models can automate biofield image analysis.
• Clustering, deep learning, and ensemble modeling offer an effective diagnostic approach.
• This system could serve as a noninvasive diagnostic support tool, pending further validation.