ULNC: Universal Language-Guided Nuclei Classification Via Instance-Aware Prompting.

Area of Science:

• Pathology
• Computer Vision
• Artificial Intelligence

Background:

• Accurate nuclei classification is vital for pathology image analysis in disease diagnosis and treatment.
• Prompt learning shows promise for universal nuclei classification by leveraging common knowledge across datasets.
• Challenges include instance variability and semantic ambiguity due to complex nuclei categories and relationships.

Purpose of the Study:

• To propose a universal language-guided nuclei classification framework (ULNC) to overcome instance variability and semantic ambiguity.
• To enhance nuclei classification at the instance level using prompt-based language supervision.
• To improve the accuracy and generalization of nuclei classification models across diverse datasets.

Main Methods:

• Developed an innovative prompt learning approach generating instance-aware text embeddings to address instance variability.
• Employed a local vision-language matching loss to reinforce semantic connections between image regions and text prompts, tackling semantic ambiguity.
• Leveraged prompt-based language supervision for nuclei classification.

Main Results:

• The proposed ULNC framework achieved superior performance compared to state-of-the-art methods.
• Demonstrated enhanced accuracy in nuclei classification across multiple public datasets.
• Showcased improved generalization capabilities to unseen domains.

Conclusions:

• ULNC effectively addresses instance variability and semantic ambiguity in nuclei classification.
• The framework shows significant potential for robust and generalized nuclei classification in diverse pathological datasets.
• ULNC advances pathology image analysis for improved disease diagnosis and treatment.