Development of Nipple Trauma Evaluation System With Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Breastfeeding Support

Background:

• No prior research exists on deep learning applications for breastfeeding support.
• Nipple trauma is a common complication affecting breastfeeding mothers.

Purpose of the Study:

• To develop a deep learning-based system for the automated evaluation of nipple trauma.
• To improve the objective assessment of nipple trauma in breastfeeding individuals.

Main Methods:

• Utilized an exploratory data analysis approach with object detection and classification algorithms.
• Developed a deep learning model using 753 images categorized by the "seven signs of nipple trauma."
• Augmented data and consolidated eight initial classes into four categories: None, Minor, Moderate, and Severe.

Main Results:

• The object detector achieved high mean average precision and Frames Per Second (FPS) for nipple and areola detection.
• The eight-class classifier demonstrated notable Area Under the Curve (AUC) values (>0.8) for specific trauma signs like fissures and scabbing.
• The four-class classifier accurately predicted severe conditions (AUC >0.7), though performance was lower for minor or unclassified categories.

Conclusions:

• A sophisticated deep learning system can automatically detect and classify nipple trauma.
• This technology has the potential to aid breastfeeding caregivers through objective image assessment.
• The system may lead to operational improvements in clinical practice and breastfeeding support.