AI Tool Spots Breastfeeding Complications from Phone Photos for Streamlined Lactation Care

Story by:

• Katherine Connor- [email protected]

Share This:

Article Content

In a step toward using artificial intelligence (AI) to provide faster and more accurate lactation care to breastfeeding mothers, engineers at the University of California San Diego developed a tool to identify breastfeeding-related conditions from simple phone images. The tool leverages neural networks to identify and classify an image of a breast as healthy, or having one of six common breastfeeding-related conditions.The goal is to use such a tool to allow lactation consultants to virtually triage patients based on the urgency of their needs and to help inform mothers when to seek professional help to avoid worsening of symptoms.

The researchers published their findings in the Journal of Medical Internet Researchon June 24, 2024.

Though the benefits of breastfeeding for both mother and infant are well documented, the World Health Organization estimates that fewer than half of infants under six months old are exclusively breastfed. Painful conditions such as nipple damage and mastitis - which affect 80 percent and 20 percent of US mothers, respectively- play a large role in a family's decision to stop breastfeeding. Access to lactation consultants who can assess these conditions and provide solutions and treatments is lacking in the US and around the globe, with even telehealth or virtual options very limited in many parts of the world.

Jessica de Souza, a graduate student in electrical and computer engineering at the University of California San Diego and first author of the study, wanted to find a way to streamline access and make this care available to more mothers in her native Brazil. After conducting surveys and focus groups with lactation consultants, de Souza learned that many lactation consultants in Brazil use the messaging platform Whatsapp to communicate with clients. She found that these professionals were overwhelmed by demand for their services, and had no way to triage who needed help urgently and who needed in-person versus remote support.

Availability of board-certified lactation consultants is limited even in the US, where there are 194 babies born per year for each available lactation consultant. In low-and middle-income countries, like Brazil, the situation is even more dire: 2.6 million babies were born in Brazil in 2021, with only 154 certified lactation consultants in the entire country. This underscores the need for better tools to improve virtual access to this care.

"While working with the lactation consultants, it was clear that breastfeeding complications were one of the most frequent reasons patients need remote urgent care. For that reason they wanted a tool that would help lactation consultants triage the level of urgency of a mom facing breast pain in their online communication platform, so they can determine the type of visit that each client needs," said de Souza. "If a mother could send a photo of her breast and have the tool to identify any specific conditions for the lactation consultant, this would help them prioritize calls or visits with mothers with more dire issues, and could one day potentially even automatically share advice or recommendations for clients with more minor complications, especially with the tools we have available today, such as virtual assistants."

In a first step toward developing such a tool, de Souza and colleagues from Professor Edward Wang's electrical and computer engineering lab in the UC San Diego Jacobs School of Engineering and UC San Diego Design Lab, studied the effectiveness of five different convolutional neural networks- a type of artificial intelligence model- in detecting whether an image of a breast depicted a healthy breast, or showed signs of one of six common breastfeeding conditions. The six conditions the model was trained to detect were abscess, mastitis, nipple blebs, dermatosis, engorgement, and nipple damage by improper feeding or misuse of breast pumps.

The researchers trained the networks using 1,000 images of breasts and nipples from breastfeeding mothers in one of these seven groups, and then tested their ability to correctly identify which condition, if any, was present in the photo. Then, they used data augmentation techniques to increase the dataset size to 6,000 images for training and validation. The best model after data augmentation was able to accurately detect healthy breasts 84.4 percent of the time, and had an average precision of 66 percent for detecting the six unhealthy breast conditions.

They then simplified the problem by having only two categories, asking the tool to identify healthy versus unhealthy breast images. They tested this binary model on 468 images, and found that the AI was able to correctly identify unhealthy breast images 95 percent of the time, and healthy images in 80.3 percent of cases. This information could be useful to lactation professionals to warn them if a patient needs more urgent attention.

De Souza says that the early findings shown in this work are promising, but there are still limitations. The researchers found that there are several image quality issues and specific breastfeeding conditions that led to poorer results. In particular, the researchers found that mastitis, engorgement, and healthy breast images are easily confused by the models due to visual similarities.

"This highlights the need for increasing the dataset to incorporate images of more varied breast complications; working closely with lactation consultants to investigate the need to categorize conditions that can be a problem but indicate false positive cases of more serious issues; and exploring the possibility of using these conditions that have higher errors as a base for following patient condition progression, where there is a transition between conditions for improving or worsening a patient's situation," the team writes.

Wang explains that a key to advancing digital health technologies for real world use is to ensure that even regular people with little to no training can still use it effectively. This is important, since a key aspect of this work is using external breast images sourced from personal devices like smartphones, instead of medical-grade images from mammograms, for example. Being able to provide effective guidelines to tell mothers exactly how to take an image on their phone will be key for optimal results from the AI.

Though it's still in the early stages and more work will be done to build a more robust image classification tool, de Souza said that this study highlights the potential to accurately detect breastfeeding complications from a phone-based app or tool.

"This type of technology will allow more people to get the care they need to successfully breastfeed," said paper co-author Kristina Chamberlain, an International Board Certified Lactation Consultant and clinical director of The Lactation Program at UC San Diego Division of Extended Studies. "Though an AI tool can't take away the personal support lactation consultants provide; it does have a place in expanding access and improving clinical outcomes for breastfeeding dyads."

Due to the impact of this work, de Souza was recently awarded a Merkin Fellowship, which will allow her to further develop her research in Brazil through a partnership between UC San Diego and the Federal University of Sao Paulo (UNIFESP), which hosts the Ana Abrão Center - Assistance, Teaching and Research in Breastfeeding and Human Milk Bank.

"The next step of this research is to leverage advanced technology to provide timely interventions that can significantly reduce the disruption caused by breastfeeding complications," said de Souza. "I am excited to expand this work in Brazil, where the need for accessible lactation care is urgent. Our goal is to help mothers continue their breastfeeding journey with confidence and support."

Paper: "Augmenting Telepostpartum Care With Vision-Based Detection of Breastfeeding-Related Conditions: Algorithm Development and Validation"

Learn more about research and education at UC San Diego in: Artificial Intelligence

Electrical engineering graduate student Jessica de Souza was motivated to find a way to streamline access to lactation care for more mothers in her native Brazil.