Automating Quality Assurance Metrics to Assess Adequate Breast Positioning in Mammography

Introduction Breast positioning is one of the key factors that can affect the performance of mammographic screening and is a leading cause of mammography units failing to earn accreditation in the United States [1]. Optimal positioning can reduce unwanted artifacts and ensure that the maximum amount of breast tissue is being included in the image. As manual quality review of mammographic images can be both subjective and time-consuming, research is underway to automate image quality assessment. Fundamental to such automated methods, is correctly identifying both the nipple and the pectoral muscle on mammographic images. Objectives The purpose of this study was to preliminarily assess whether breast positioning for mammography might be objectively analyzed. The first objective of this study was to assess the performance of a novel algorithm for automated nipple detection. The second objective was to assess two additional breast positioning metrics that rely on correct identification of the nipple (i.e. the relative lengths of the posterior nipple line on the CC and MLO views and the extension of the pectoral muscle relative to the position of the nipple on the MLO) and how these metric compare to a visual clinical image review and automated breast volume estimates. Materials & Methods The dataset used to evaluate the breast-positioning algorithm comprised of standard four–view mammographic studies [i.e. left craniocaudal (LCC), right CC (RCC), left mediolateral oblique (LMLO) and right MLO (RMLO)] from 23 women. The algorithm was run over the raw images to automatically assess the following:  correct identification of the nipple that the length of the posterior nipple line (PNL) on the CC views was within 1 cm of that on the MLO views  that the inferior edge of the pectoral muscle on the MLO views extends to or below the level of the nipple. The results of the algorithm were visually assessed to determine its accuracy. Comparisons of the breast volume estimates between CC and MLO views and left and right breasts were used as a surrogate measure of the consistency in breast positioning. Conclusion Automated methods for assessing adequate breast positioning in mammography have the potential to improve the quality of mammographic images and, therefore, clinical performance. Automating Quality Assurance Metrics to Assess Adequate Breast Positioning in Mammography The mortality benefits of mammographic screening are largely dependent on the ability to detect cancers at as early stage as possible. Therefore, it is critical that mammograms are acquired and interpreted to the highest possible standards of quality. Efforts to ensure and improve the quality of screening mammography (including breast tomosynthesis) should focus on aspects of the acquisition process that are known to affect the sensitivity of the final interpretation. Breast positioning is one of the most important of these factors and is a leading cause of accreditation failure in the United States. 1 More importantly, poor positioning can lead directly to delayed diagnosis of breast cancer, potentially impacting patient mortality and morbidity. With delayed diagnosis of breast cancer also retaining its position as the number one source of medical legal risk, 2 the impact of poor positioning goes far beyond its importance as an accreditation standard. Although positioning plays a critically important role in image quality, automated tools to assess breast positioning in real-time are currently lacking. There are numerous clinical benefits to presenting a set of easily understandable, objective metrics of breast positioning, including:  The potential for technologists to retake inadequately positioned images immediately, while the patient is still in the clinic, rather than having to re-schedule or recall her.  The opportunity for incorporating automated positioning metrics into quality assurance guidelines.  The ability to perform system-wide and population-wide analyses comparing breast positioning. In this study, the performance of a novel algorithm for the automatic assessment of two breast positioning metrics from digital mammograms was assessed by comparison to a visual clinical review and automated, volumetric, breast density measurements.