Cervical Intraepithelial Neoplasia Grading from Prepared Digital Histology Images.

Area of Science:

• Digital pathology
• Medical imaging analysis
• Artificial intelligence in oncology

Background:

• Cervical intraepithelial neoplasia (CIN) diagnosis relies on histopathology.
• Automating CIN grading from digital histology images is crucial for efficient cancer screening.
• Current methods require expert interpretation, leading to potential variability.

Purpose of the Study:

• To develop and evaluate an integrated deep learning approach for automated CIN diagnosis.
• To identify the optimal convolutional neural network (CNN) architecture for classifying epithelial patches.
• To fuse patch-level predictions for accurate histology sample grading.

Main Methods:

• Seven CNN architectures were evaluated for classifying epithelial patches.
• The best performing CNN was selected for further analysis.
• Three fusion methods were applied to combine patch predictions.
• An ensemble model integrated the CNN classifier and the best fusion method.

Main Results:

• The model ensemble achieved a diagnostic accuracy of 94.57%.
• This performance surpasses existing state-of-the-art classifiers for cervical cancer histopathology.
• The study identified optimal deep learning models and fusion techniques for CIN diagnosis.

Conclusions:

• The proposed integrated approach demonstrates high accuracy in automated CIN diagnosis.
• This work advances automated analysis of digital histopathology images for cervical cancer.
• Further research in automated CIN diagnosis can benefit from this methodology.