Deep-Optimal Leucorrhea Detection Through Fluorescent Benchmark Data Analysis

Area of Science:

• Medical Diagnostics
• Computational Biology
• Biomedical Imaging

Background:

• Vaginitis, a common gynecological condition, requires accurate diagnosis for effective treatment.
• Current diagnostic methods like wet mounts and Gram staining have limitations in precision.
• Fluorescent staining offers enhanced visualization of vaginal components.

Purpose of the Study:

• To develop an advanced diagnostic tool for vaginitis using deep learning and fluorescent staining.
• To create a comprehensive dataset for training and evaluating AI models for leucorrhea detection.
• To present a novel, lightweight deep learning network (LRNet) for efficient and accurate vaginitis diagnosis.

Main Methods:

• Established a large-scale dataset (343K labels) of multiple fluorescence leucorrhea images across 8 categories.
• Developed LRNet, a lightweight deep learning network featuring Ghost modules and deformable convolutions.
• Utilized fluorescent staining for distinct visualization of cellular and pathogenic elements in vaginal discharge.

Main Results:

• The LRNet model demonstrated superior performance over conventional detection networks.
• LRNet achieved significant reductions in model parameters (up to 91.4%) and FLOPs (74%).
• The network effectively detects crucial indicators for vaginal health.

Conclusions:

• LRNet offers a robust and efficient solution for diagnosing vaginitis.
• The proposed method enhances the precision and speed of identifying vaginal health indicators.
• This approach has the potential to significantly improve clinical diagnostic capabilities for vaginitis.