Convolutional neural network-based computer-aided diagnosis in Hiesho (cold sensation)

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Hiesho (cold sensation) is a prevalent global health issue predominantly affecting women, characterized by cold extremities and potential links to chronic diseases.
• Current Hiesho diagnosis relies on subjective methods like questionnaires, leading to diagnostic controversies and increased patient/doctor burden.
• Previous research indicated significant temperature differences between the forehead and plantar soles in Hiesho patients.

Purpose of the Study:

• To develop and evaluate a quantitative, automatic computer-aided diagnosis (CAD) system for Hiesho using convolutional neural networks (CNNs).
• To assess the feasibility of using thermographic images for objective Hiesho diagnosis.
• To improve diagnostic accuracy and reduce the subjective nature of current Hiesho assessment methods.

Main Methods:

• A CNN model, specifically AlexNet, was trained using 5612 thermographic images from 46 participants (23 Hiesho patients, 23 healthy controls).
• The performance of the CNN-based CAD system was evaluated against other machine learning models.
• Key performance metrics included accuracy, precision, sensitivity, specificity, and F1 score.

Main Results:

• The proposed CNN-based Hiesho CAD system achieved a perfect performance score of 100% across all evaluated metrics (accuracy, precision, sensitivity, specificity, F1 score).
• Thermographic imaging demonstrated high feasibility in distinguishing between Hiesho patients and healthy individuals.
• The CNN-based CAD system proved highly accurate and reliable for the automatic diagnosis of Hiesho.

Conclusions:

• Thermographic imaging is a viable tool for objective Hiesho discrimination.
• CNN-based CAD systems offer a highly accurate and reliable method for the automatic diagnosis of Hiesho.
• This approach has the potential to significantly enhance Hiesho diagnosis, benefiting both patients and healthcare providers.