Key points are not available for this paper at this time.
This study addresses the development of EndoClassify, an artificial intelligence (AI) model designed to assess endometrial characteristics and enhance embryo receptivity. Utilizing a dataset of 402 endometrial ultrasound images augmented to 14.989, EndoClassify, incorporating Attention U-Net for image segmentation and GoogLeNet Inception for image classification, demonstrated exceptional performance with an accuracy of 95%, loss of 10%, a sensitivity of 93%, and specificity of 93%. The significance of EndoClassify extends beyond its robust metrics. This AI model has transformative potential in clinical settings, offering specialists a reliable, rapid, and accurate tool for endometrial assessment in assisted reproduction technology (ART) cycles. Identifying ‘good endometrium’ with 71% accuracy, corresponding to a 74% pregnancy rate, underscores EndoClassify’s role in significantly improving patient outcomes. In conclusion, the seamless integration of ultrasonographic parameters and AI techniques enhances efficiency in clinical decision-making and signifies a crucial collaboration between advanced technology and clinical expertise. While acknowledging the retrospective design as a limitation, it is imperative to highlight potential biases introduced by this design. Additionally, including fresh and frozen embryo transfers without known ploidy status adds transparency to the study’s limitations. EndoClassify stands as a beacon of progress, poised to revolutionize personalized treatment strategies and bring tangible benefits to specialists and patients in the dynamic landscape of assisted reproductive technology.
Asch-Schuff et al. (Wed,) studied this question.