NAS-SGAN: A Semi-Supervised Generative Adversarial Network Model for Atypia Scoring of Breast Cancer...

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Medical image analysis

Background:

• Nuclear atypia scoring (NAS) is crucial for breast cancer prognosis and treatment planning.
• Manual grading of histopathological images is subjective and labor-intensive.
• Automated grading requires large annotated datasets, which are scarce and costly.

Purpose of the Study:

• To develop a robust automated system for breast cancer grading using limited annotated data.
• To improve the accuracy and efficiency of cancer diagnosis through quantitative image analysis.
• To explore the application of semi-supervised generative adversarial networks (GANs) for histopathological image classification.

Main Methods:

• Proposed a novel semi-supervised generative adversarial network (NAS-SGAN) model.
• Employed adversarial training with both labeled and unlabeled histopathological samples.
• Integrated unsupervised and supervised learning within the discriminator, sharing model parameters.
• Utilized a stable feature matching objective function for generator training.

Main Results:

• The NAS-SGAN model demonstrated improved discrimination between different cancer grades.
• Enhanced robustness and accuracy of the automated grading system were achieved.
• Effective performance was validated even with a limited amount of labeled data.

Conclusions:

• Semi-supervised GANs offer a promising approach for automated cancer grading with limited annotated data.
• The NAS-SGAN model can assist pathologists by providing accurate and efficient cancer diagnosis.
• This framework has the potential to significantly improve individualized treatment planning and disease prognosis.