Category
Applied
Description
The present study investigates the ability of computer vision technology to count primordial follicles in the primate ovary. The number of primordial follicles, i.e., ovarian reserve, reflects the ovary’s potential of supporting endocrine and reproductive function in females. Follicle counting has been used in translational and clinical studies to assess ovarian health and response to regulatory factors or toxic substances. This critical process is currently performed manually by experienced laboratory personnel. Existing research on computer vision software’s ability to count follicles has yielded limited results. Therefore, in reaction to the labor-intensive, time-consuming nature of manual follicle counting and technological growth in the field of computer vision, experiments were designed to answer the question: “How accurately can novel computer vision models count primordial follicles in the primate ovary?” To answer this question, rhesus monkey ovarian sections stained with hematoxylin and eosin were photographed. Trained human counters counted and annotated primordial follicles (n = 808) using Computer Vision Annotation Tool (CVAT). The annotations were exported and translated into a model-readable data format, which was then given to three different computer vision models, CellposeSAM, Detectron2, and YOLO. Each model’s parameters were tuned by three different two-person sub-teams to yield optimal results. At present, the best performing model, CellposeSAM, achieves an 86% F1-score with 78% precision and 97% recall. This research is limited to primordial follicle counting in the nonhuman primate ovary. Future research could explore counting follicles at advanced developmental stages and apply the current models to human follicle counting. In addition, a larger pool of data will achieve more accurate results.
Computer Vision for Ovarian Follicle Analysis
Applied
The present study investigates the ability of computer vision technology to count primordial follicles in the primate ovary. The number of primordial follicles, i.e., ovarian reserve, reflects the ovary’s potential of supporting endocrine and reproductive function in females. Follicle counting has been used in translational and clinical studies to assess ovarian health and response to regulatory factors or toxic substances. This critical process is currently performed manually by experienced laboratory personnel. Existing research on computer vision software’s ability to count follicles has yielded limited results. Therefore, in reaction to the labor-intensive, time-consuming nature of manual follicle counting and technological growth in the field of computer vision, experiments were designed to answer the question: “How accurately can novel computer vision models count primordial follicles in the primate ovary?” To answer this question, rhesus monkey ovarian sections stained with hematoxylin and eosin were photographed. Trained human counters counted and annotated primordial follicles (n = 808) using Computer Vision Annotation Tool (CVAT). The annotations were exported and translated into a model-readable data format, which was then given to three different computer vision models, CellposeSAM, Detectron2, and YOLO. Each model’s parameters were tuned by three different two-person sub-teams to yield optimal results. At present, the best performing model, CellposeSAM, achieves an 86% F1-score with 78% precision and 97% recall. This research is limited to primordial follicle counting in the nonhuman primate ovary. Future research could explore counting follicles at advanced developmental stages and apply the current models to human follicle counting. In addition, a larger pool of data will achieve more accurate results.
