CaMeL-Net: Centroid-aware metric learning for efficient multi-class cancer classification in pathology images

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Image analysis

Background:

• Cancer grading is crucial for patient care, treatment, and management.
• Artificial neural networks show promise for improving cancer diagnosis accuracy.
• Current deep learning models often require significant computational resources.

Purpose of the Study:

• To propose an efficient convolutional neural network (CNN) for accurate and robust multi-class cancer classification.
• To leverage metric learning for improved performance in pathology image analysis.
• To develop a computationally efficient model for cancer grading.

Main Methods:

• A centroid-aware metric learning network is proposed for pathology image analysis.
• The network optimizes relative distances between image features using class centroids.
• A novel loss function and training strategy are introduced for enhanced optimization.

Main Results:

• The proposed method achieved high accuracy (e.g., 88.7% for colorectal cancer) and F1-scores across multiple datasets.
• Performance metrics include accuracy, F1-score, and quadratic weighted kappa.
• The model demonstrated superior performance and computational efficiency compared to competing methods.

Conclusions:

• The developed network provides accurate and reliable cancer classification predictions.
• The method achieves superior computational efficiency in both training and inference.
• Future work includes further development and application to other domains.