Prospective Ultrasound Data Collection Study for AI-Based Detection of Breast Biopsy Markers

Overview

The goal of this observational study is to collect ultrasound imaging data to support the development and refinement of an artificial intelligence (AI)-enhanced ultrasound software designed to detect and localize standard-of-care breast biopsy markers in adult female patients undergoing surgical excision of a breast tumor. The main questions it aims to answer are: Can the study software identify the location of previously placed breast biopsy markers during routine ultrasound examinations? Participants will undergo limited breast ultrasound imaging during a standard-of-care scheduled clinical visit. During the ultrasound imaging, the ultrasound system will automatically record the ultrasound images for the development and training of the AI software The study does not change standard medical care, and no diagnostic or treatment decisions are made based on the study software.

Study Design and Setting This is a prospective, non-interventional observational study designed to collect ultrasound imaging data and related metadata to support the development, refinement, and performance characterization of an artificial intelligence (AI)-enhanced ultrasound software intended to detect and localize standard-of-care breast biopsy markers. All ultrasound examinations are performed as part of routine clinical care for patients undergoing evaluation and planned surgical excision of a breast tumor. Study participation does not alter standard imaging protocols, clinical workflow, or treatment decisions. Study Procedures During a scheduled clinical visit, participants undergo limited B-mode breast ultrasound imaging using an FDA-cleared ultrasound system provided for the study. The system includes investigational software that displays marker localization output during image acquisition. The software output is not used to guide or alter clinical decision-making. Ultrasound image data are automatically recorded by the system during the examination. Imaging datasets include visualization of the biopsy marker and surrounding breast tissue. When applicable, imaging may occur at different time points along the surgical treatment pathway. A brief user-experience questionnaire may be completed by the operating physician following image acquisition. De-identified clinical and imaging metadata are recorded in a structured case report form (CRF). Data Collection and Management Ultrasound image data are de-identified prior to transfer. De-identified datasets are securely transmitted to the sponsor's controlled research environment for algorithm development and analysis. Collected variables may include imaging parameters, marker type, breast tissue characteristics, tumor characteristics, and operator experience level. A structured data dictionary defines all variables collected, including source, format, allowable ranges, and coding conventions where applicable. All study data are stored in compliance with applicable data protection and privacy regulations. Access to study data is restricted to authorized personnel. Quality Assurance and Monitoring Data quality procedures include predefined validation checks for completeness, internal consistency, and logical range limits within the electronic data capture system. Automated and manual data review processes are implemented to identify missing, inconsistent, or out-of-range values. Source data verification may be performed on a risk-based sample of records to confirm consistency between recorded study data and source documents, such as medical records or ultrasound system exports. Standard Operating Procedures (SOPs) govern study conduct, including patient recruitment, data acquisition, data transfer, data management, adverse event documentation, and change management. Monitoring activities may be conducted according to a risk-based plan. Statistical Considerations This study is primarily descriptive and developmental in nature. Analyses will focus on characterization of software marker localization performance, including measures such as detection success, time to localization, and performance across relevant subgroups (e.g., tissue characteristics, operator experience levels). Agreement analyses may be performed comparing software output to trained reader assessments of recorded ultrasound images. Statistical methods may include descriptive statistics, confidence interval estimation, and exploratory subgroup analyses. The study is not designed as a confirmatory clinical effectiveness trial. Sample Size Rationale The planned enrollment is based on feasibility considerations and the need to collect a sufficient number of imaging datasets to support algorithm development and performance characterization across a range of clinical and anatomical conditions. The study is not powered for confirmatory hypothesis testing.