Additional file 5 of Micro-CT and machine learning: a high-throughput alternative to histology for follicle reserve assessment in cryopreserved ovarian tissue

Additional file 5. Additional Table 1. Comparison of segmentation performance metrics (DICE, Precision, Recall) between U-net model output and Avizo post-processing. DICE scores were generally higher following Avizo post-processing, with most samples exceeding the acceptable threshold of 0.7. Samples C1 and A3 showed comparable values between methods. Precision values consistently improved with post-processing. In contrast, Recall values tended to decrease due to a deliberate pixel-level shrinkage applied during post-processing to prevent 3D oocyte model merging. This shrinkage, typically one pixel layer, minimized redundancy and preserved spatial distribution. Sample A2 was excluded from the table due to the absence of detectable oocytes. Despite the trade-off in size accuracy, Avizo post-processing was retained in the pipeline to prioritize reliable oocyte count and localization.