Assessment of Oocyte Quality Using Deep Radiomic Signature of Oocyte Morphology in the Mouse

Abstract

Background: Technologies such as in vitro fertilization (IVF) are increasingly used to compensate for the loss of fertility associated with increasing maternal age and decline in oocyte quality. During IVF, superovulation yields more oocytes than are required, and the selection of oocytes for fertilization and implantation is based on poorly defined morphological features which are subject to user bias. Aim: We aim to develop a deep radiomic signature based on an artificial intelligence (AI) model in order to identify oocyte morphological changes corresponding to reproductive ageing in bright field images captured by optical light microscopy. This approach has potential for application in assessment of human oocytes. Method: Oocytes were collected from three groups of C57BL/6J female mice: young (4- to 5-week-old), aged (12-month-old) mice and aged mice treated with the NAD+ precursor nicotinamide mononucleotide (NMN), a treatment recently shown to rejuvenate aspects of fertility in aged mice. Deep learning, swarm intelligence and discriminative analysis were applied to images of mouse oocytes taken by bright field microscopy to identify a highly informative deep radiomic signature (DRS) of oocyte morphology. Approximately 25 oocyte images were derived from each group. Results: This signature distinguished morphological changes in oocytes associated with maternal ageing, which are imperceptible to an experienced embryologist, with 92% accuracy (AUC[Formula: see text]1), reflecting the age-induced decline in oocyte quality. We then employed the DRS to evaluate the impact of the NMN treatment. The DRS signature classified 60% of oocytes from NMN-treated aged mice as having a ‘young’ morphology, demonstrating the signature’s sensitivity to improvements in quality and reinforcing its applicability for oocyte selection. Conclusion: Our data illustrate the power of DRS for recognizing morphological features of cellular ageing, which outperforms current subjective methods relying on visual grading for the recognition and classification of oocytes based on maternal age.