A Partial Label-Based Machine Learning Approach For Cervical Whole-Slide Image Classification: The Winning TissueNet...

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Cervical cancer is a major global health concern for women.
• Accurate classification of cervical intraepithelial lesions is crucial for timely treatment.
• Pathologist analysis of biopsies is time-consuming and prone to variability.

Purpose of the Study:

• To develop an automated pipeline for classifying cervical lesion status from whole slide images (WSIs).
• To present the winning solution from the TissueNet Challenge for cervical slide classification.
• To introduce a novel partial label-based loss function for improved WSI analysis.

Main Methods:

• A two-step classification model was employed, starting with an ensemble CNN for patch classification.
• Support Vector Machine (SVM) was used for slide-level classification based on aggregated patch predictions.
• A novel partial label-based loss function was introduced to leverage weakly supervised data.

Main Results:

• The proposed method achieved a winning score of 94.7% in the TissueNet Challenge.
• The approach successfully classified cervical lesion status from WSIs.
• The novel loss function enhanced performance without requiring additional expert annotations.

Conclusions:

• The developed AI algorithm offers an efficient and accurate method for automated cervical cancer classification.
• This approach can reduce diagnostic variability and time in digital pathology.
• The findings pave the way for integrating AI tools in cervical cancer treatment planning.