Abstract

This paper presents a novel convolutional neural network architecture specifically designed for automated detection and classification of malignant cells in whole slide histopathology images. Our method achieves state-of-the-art performance on multiple public datasets and demonstrates clinical applicability in high-throughput pathology labs.

Key Contributions

1. Novel attention mechanism for multi-scale cell detection
2. Efficient processing of gigapixel WSI images
3. Clinical validation across multiple institutions
4. Publicly available model and code

Results

• Sensitivity: 94.2% at 95% specificity
• Computational Time: 3.5 minutes per whole slide image
• Cross-institutional Validation: AUC of 0.92

See full publication for more details.