Dataset on automation bias in AI-assisted breast ultrasound diagnosis: 295 static grayscale ultrasound images, pathological standards, and multireader multicase reading records from 148 readers

This dataset supports the article "Evaluation of Automation Bias in Breast Ultrasound Image with Different Assistance Scenarios of Artificial Intelligence: A Multireader Multicase Study" and was compiled to investigate automation bias in artificial intelligence (AI) assisted breast ultrasound diagnosis.Data were retrospectively collected from three tertiary referral hospitals in southern, central, and northern China between January 2021 and December 2023.

Eligible participants were female patients aged 18 years or older who underwent breast ultrasound examination and had lesions categorized as BI-RADS 3 to 5. After independent review by two radiologists with more than three years of experience, 295 eligible cases were included, each contributing one static grayscale ultrasound image showing the largest cross section of the target lesion. The reference standard was core needle biopsy or surgical pathology for malignant cases, and pathology or stable imaging surveillance for more than three years for benign cases.The dataset comprises three components: (1) 295 de-identified static grayscale breast ultrasound images in JPG format; (2) a table of pathological or follow-up reference standards matched one to one with the images, including benign or malignant diagnosis and BI-RADS category; (3) a multireader multicase (MRMC) reading record in CSV format, documenting the diagnostic decisions of 148 radiologists from 89 hospitals across China on 80 cases (selected from the 295 cases by BI-RADS stratified random sampling) under three reading scenarios: unaided (Scenario A), simultaneous AI-aided (Scenario B), and sequential AI-aided (Scenario C).