A Chain-of-thought Reasoning Breast Ultrasound Dataset Covering All Histopathology Categories

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Breast ultrasound (BUS) is crucial for diagnosing breast lesions, but AI development faces limited, low-richness datasets.
• Existing BUS datasets lack the scale and detailed annotations needed for advanced AI reasoning.

Purpose of the Study:

• To introduce BUS-CoT, a comprehensive breast ultrasound dataset designed for chain-of-thought (CoT) reasoning analysis in AI.
• To provide a high-quality benchmark dataset for training and evaluating AI models in breast lesion diagnosis, including rare cases.

Main Methods:

• Developed BUS-CoT with 11,439 ultrasound images covering 11,850 lesions across all 99 WHO histopathology categories.
• Curated a high-quality subset of 5,163 lesion-focused images annotated by expert radiologists.
• Constructed CoT reasoning processes (observation, feature, diagnosis, pathology) verified by experts.

Main Results:

• The BUS-CoT dataset offers extensive data scale and annotation richness for AI development.
• Includes detailed reasoning pathways to facilitate CoT analysis and improve diagnostic accuracy.
• Covers all histopathology types, aiding AI robustness in diagnosing rare and challenging breast lesions.

Conclusions:

• BUS-CoT addresses the limitations of current BUS datasets, enabling advanced AI research in diagnostic reasoning.
• The dataset facilitates the development of more robust and accurate AI systems for breast lesion classification, especially for rare conditions.
• Public availability of the dataset and code promotes further research and development in AI-assisted breast imaging.