LesionDetect

Breastfeeding offers numerous long-term benefits for mother and baby. However, mothers may face challenges such as low milk supply, fatigue, medical problems, difficulties with feeding techniques or pain, and lack of social support. Regarding pain and complications while breastfeeding, it is estimated that about 80% of mothers might experience nipple pain and fissures. These complications are painful and threaten breastfeeding continuation since some lesions, when not correctly identified and treated, might take longer to treat and can cause breastfeeding interruption. Offering tools to professionals to appropriately triage and diagnose breast and nipple lesions can streamline lactation care and allow faster treatment of these lesions. Artificial intelligence (AI) can benefit nurses and lactation specialists, as currently there are models available to identify breastfeeding complications. 

In this work, we inform the development of an AI-based tool that helps identify nipple injuries caused by breastfeeding complications. This tool focuses on identifying nipple injuries based on the Nipple and Areola Complex Lesions Classification Instrument (ILMA) and the Nipple Trauma Score (NTS), providing a comprehensive overview of the clinical condition of a nipple injury and helping healthcare providers find the most suitable treatment based on the severity of the lesion. We aim to assess the applicability of this tool as an auxiliary resource for nurses and lactation professionals to identify painful nipple lesions quickly, to better triage their patients, and enhance the overall experience and care for breastfeeding mothers. 

We conducted an ethnographic study involving 10 hours of observations of lactation specialists at a milk bank in Brazil across two scenarios: (1) during clinical group discussions on breastfeeding cases with nipple lesions and (2) during a retrospective evaluation of 900 images of nipple lesions by two internationally board-certified lactation specialists (IBCLCs). These observations provided in-depth insights into the specialists’ workflows and decision-making processes. The data collected from these sessions informed the design of a tool to support clinical decision-making and case evaluation for healthcare providers. 

Through observations and iterations with the lactation specialists, we identified tools used to assess nipple lesions and trauma levels. Using these insights, we designed a web-based clinical decision tool incorporating user interface (UI) components and two deep-learning models to evaluate the ILMA and NTS measurements. In its initial phase, the tool features the following functionalities: an image selector for uploading images for the analysis, navigation buttons to change between images and a save button to record the analysis, an AI detection tool that estimates the type and severity of the lesion with a confidence score, manual selection of nipple color and wound location, manual ILMA and NTS input based on the clinician’s own evaluation, area calculation tool to assess the proportion of the wound relative to the nipple’s area, increasing accuracy on the NTS score, note-taking section to record observations about each image. 

This work introduced a diagnostic tool to assist nurses and lactation specialists in identifying and classifying nipple lesions caused by breastfeeding. In addition to facilitating clinical discussions on nipple trauma treatments, the tool offers further applications, such as serving as a valuable training resource for lactation care students. Additionally, it can function as a labeling tool, contributing to creating more standardized datasets for future AI modeling and enhancing the development of automated systems for diagnosing breastfeeding complications.