Cervical Lesion Classification Method Based on Cross-Validation Decision Fusion Method of Vision Transformer and...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Cervical cancer diagnosis relies heavily on accurate image analysis.
• Existing diagnostic methods may have limitations in sensitivity and specificity.
• Integrating advanced AI models can potentially improve diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a cross-validation decision-making fusion method for enhanced cervical lesion classification.
• To adapt Vision Transformer and DenseNet161 models for clinical application in acetic acid imaging.
• To improve the accuracy and reduce missed detection rates in cervical lesion diagnosis.

Main Methods:

• Utilized a critical dataset of acetic acid images for clinical diagnosis.
• Applied a fivefold cross-validation method to train Vision Transformer and DenseNet161 models.
• Fused prediction results from both models using weighted averaging for final classification.

Main Results:

• The proposed fusion method achieved an accuracy rate of 68% for classifying four types of cervical lesions.
• Demonstrated the effectiveness of combining deep learning models for medical image analysis.
• Indicated a reduction in missed detection rates compared to individual models.

Conclusions:

• The developed fusion method is suitable for clinical environments, improving diagnostic efficiency.
• The approach effectively reduces the missed detection rate of cervical lesions.
• Enhances patient care and health outcomes by ensuring timely and accurate diagnosis.