The utility of artificial intelligence and deep learning to automate and accelerate follicle counts in human ovarian tissue†

Abstract

Follicles comprised of oocytes and surrounding cells are essential for reproductive function. They are fixed before birth and decrease thereafter through a process of activation, growth, and apoptosis. The advent of ovarian tissue cryopreservation is critical for fertility preservation and has allowed us to evaluate follicle numbers and folliculogenesis in humans. However, current histopathological assessments are labor intensive and subject to interobserver variability. We developed an AI-based method that integrates deep learning segmentation and object detection to automate follicle counting in whole-slide images (WSIs) of ovarian tissue. Using 1857 WSIs from 47 patients and 8300 annotated follicles, our method employs DeepLabV3+ for segmentation and Faster R-CNN for object detection. Predictions from both models are merged, and performance metrics (Dice coefficient, sensitivity, and positive predictive value) were calculated pre- and post-failure analysis. The segmentation model achieved a Dice coefficient of 0.4939, while the object detection model achieved a COCOmetric score 0.27. The merged results of both models performed with a sensitivity of 0.92 and PPV of 0.95, after manual correction of annotations. Our AI-driven approach enhances follicle quantitation accuracy and reproducibility, representing a promising tool to support research and clinical decision-making in fertility preservation. Artificial intelligence models effectively automate the labor-intensive process of ovarian follicle counting, improving accuracy and efficiency, which could improve outcomes in ovarian tissue transplantation after cryopreservation and future fertility success.