Visualizing ambiguity in semantic segmentation of ICSI support systems for polar-body detection

Intracytoplasmic sperm injection (ICSI) is an assisted reproductive technology that involves injecting sperm into an oocyte using micromanipulation pipettes. While the demand for ICSI is constantly increasing, there is a shortage of skilled embryologists to perform this intricate procedure. Therefore, an assistance system that can enable even novice operators to perform ICSI successfully is required. One of the key challenges for novices is the accurate identification of polar bodies within the oocyte. This paper proposes an image-presentation system for assisting novice operators in recognizing polar bodies during ICSI by highlighting them using semantic segmentation. To foster operator trust, the system deliberately represents the ambiguity in its inference using Monte Carlo dropout. The effectiveness of the proposed system was validated through oocyte rotation experiments conducted with 18 novice operators. The results indicated that the proposed method enhanced the usability of micromanipulation systems for ICSI.