Comparing Deep Learning Models for Multi-cell Classification in Liquid- based Cervical Cytology Image.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Liquid-based cytology (LBC) is a standard automated method for Papanicolaou (Pap) smear screening.
• Accurate cell segmentation is a critical yet challenging prerequisite for current LBC analysis methods.
• Factors like stain variability and cell clustering complicate automated analysis.

Purpose of the Study:

• To develop an automated method for cervical slide image classification.
• To overcome limitations of existing cell segmentation-dependent approaches.
• To enable accurate cellular-level evaluation for improved cervical cancer screening.

Main Methods:

• Generation of labeled cervical cell image patches.
• Fine-tuning Convolutional Neural Networks (CNNs) for deep hierarchical feature extraction.
• Implementation of a novel graph-based approach for cell detection and evaluation.

Main Results:

• The proposed pipeline successfully classifies cervical images containing both single and overlapping cells.
• The VGG-19 model demonstrated superior performance in classifying cervical cytology patch data.
• Achieved 95% accuracy for cervical cytology patch classification based on the precision-recall curve.

Conclusions:

• The developed method offers a robust pipeline for automated cervical slide image analysis.
• The approach enhances the reliability of liquid-based cytology screening by addressing segmentation challenges.
• Deep learning and graph-based methods show significant promise for advancing cytopathology automation.