Global context-aware cervical cell detection with soft scale anchor matching.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Manual cervical cancer screening is labor-intensive and prone to errors.
• Automated methods using convolutional neural networks reduce reliance on manual feature extraction but often produce false positives.
• Accurate cell segmentation is a traditional bottleneck in automated cervical cell analysis.

Purpose of the Study:

• To develop a global context-aware framework for automated cervical cancer screening.
• To reduce false positive predictions in cell detection without requiring segmentation.
• To improve the overall performance and efficiency of computer-aided screening systems.

Main Methods:

• A global context-aware framework integrating image-level classification and weighted loss to filter false positives.
• A novel soft scale anchor matching strategy for improved ground truth assignment across feature pyramid scales.
• Elimination of the need for explicit cervical cell segmentation.

Main Results:

• Achieved a 5.7% increase in mean average precision.
• Improved specificity by 18.5% compared to existing methods.
• Introduced a minor 2.6% delay in inference time.

Conclusions:

• The proposed framework effectively reduces false positives in automated cervical cancer screening.
• Avoiding cell segmentation simplifies the process and reduces workload for pathologists.
• Demonstrates significant potential for automation-assisted cervical cancer diagnosis.