A Data-Efficient Framework for the Identification of Vaginitis Based on Deep Learning.

Area of Science:

• Gynecological health
• Medical imaging analysis
• Artificial intelligence in diagnostics

Background:

• Vaginitis diagnosis traditionally relies on time-consuming manual microscopy.
• Deep learning offers automated diagnosis but requires extensive annotated data.
• Manual annotation of microscopic images is costly and requires expert personnel.

Purpose of the Study:

• To develop a data-efficient framework for vaginitis identification.
• To reduce the high cost associated with labeling microscopic images for deep learning models.
• To adapt active learning strategies for complex microscopic image analysis.

Main Methods:

• A novel framework combining transfer learning and active learning strategies was proposed.
• An informative sample selection strategy was employed to identify a minimal training subset.
• A pre-trained convolutional neural network (CNN) was fine-tuned on the selected data subset.

Main Results:

• The proposed pipeline achieved competitive performance in vaginitis diagnosis.
• The framework demonstrated a significant reduction in annotation costs, saving 37.5%.
• The approach proved effective despite the complex nature of microscopic images.

Conclusions:

• The developed framework offers a cost-effective solution for diagnosing vaginitis from microscopic images.
• This approach significantly reduces the financial burden of data annotation.
• The framework shows potential for broad application in other microscopic imaging fields, including blood analysis.