Abstract
Background/Objectives: Nipple trauma is a common challenge during the early postpartum period, often undermining maternal confidence and breastfeeding success. Although deep-learning-based image analysis offers the potential for objective and remote assessments, its feasibility in clinical practice has not been well examined. This study aimed to evaluate the feasibility and acceptability of a deep-learning-based nipple trauma assessment system and explore maternal perceptions of the intervention. Methods: A quasi-experimental study was conducted at a maternity hospital in Japan. Participants were assigned to intervention or control groups based on their delivery month. Mothers in the intervention group used a dedicated offline smartphone to photograph their nipples during hospitalization. Images were analyzed using a pretrained deep-learning model, and individualized feedback was delivered via a secure messaging platform. Self-administered questionnaires were collected at three points: late pregnancy, during hospitalization, and one month postpartum. Maternal experiences and satisfaction with breastfeeding were also assessed. Results: A total of 23 participants (intervention = 8 and control = 15) completed the study. The system functioned without technical errors, and no adverse events were reported. Most participants found the AI results useful, with 75% receiving high-confidence outputs (predicted class probability ≥ 60%). Participants expressed interest in real-time feedback and post-discharge use. Breastfeeding self-efficacy scores (BSES-SF) improved more in the intervention group (+9.8) than in the control group (+7.8). Conclusions: This study confirmed the feasibility and acceptability of a deep-learning-based nipple trauma assessment system during postpartum hospitalization. The system operated safely and was well received by participants. Future developments should prioritize real-time, remote functionality to support diverse maternal needs.