Artificial intelligence in polycystic ovarian syndrome management: past, present, and future

Abstract

Background: Integrating artificial intelligence (AI) prospected in the practical clinical management of polycystic ovary syndrome (PCOS) promised significant improvement in efficiency, interpretability, and generalizability.

Purpose: To delineate a comprehensive inventory of AI-driven interventions pertinent to PCOS across diverse clinical contexts.

Evidence reviews: AI-based analytics profoundly transformed the management of PCOS, particularly in the domains of prediction, diagnosis, classification, and screening of potential complications.

Results: Our analysis traced the principal applications of AI in PCOS management, focusing on prediction, diagnosis, classification, and screening. Furthermore, this study ventures into the potential of amalgamating and augmenting existing digital health technologies to forge an AI-augmented digital healthcare ecosystem encompassing the prevention and holistic management of PCOS. We also discuss strategic avenues that may facilitate the clinical translation of these innovative systems.

Conclusion: This systematic review consolidated the latest advancements in AI-driven PCOS management encompassing prediction, diagnosis, classification, and screening of potential complications, developing a digital healthcare framework tailored to the practical clinical management of PCOS.

Keywords: Artificial intelligence; Digital healthcare; Polycystic ovary syndrome.

Fig. 1

Selection process of the studies. Article selection flow chart for studies related to AI and PCOS according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines

Fig. 2

Current applications of machine intelligence in PCOS management. Common algorithms used in supervised learning include (1) artificial neural networks, such as CNN, KNN, PNN, and DNN; (2) Bayesian learning, such as NB; (3) DT, such as classification and regression tree, and supervised learning in quest; (4) ensemble methods, such as RF, and XGBoost; and (5) linear models, such as linear regression, LR, GLM, SVM. Common algorithms used in unsupervised learning include (1) clustering, such as k-means and GMM; and (2) dimensionality reduction, such as SOM

Fig. 3

Overview AI implementations in PCOS management based on prevention strategies

Fig. 4

The framework of federated learning. According to the data partition, it’s mainly classified into horizontal and vertical federation learning models. The horizontal federation learning model applies to the condition where different users in the same domain have similar characteristics. Local hospitals got copies of the current global model from a federated server to train on their datasets, sending the model updates back to the federated server with a certain number of iterations, keeping their datasets in their secure infrastructure. The federated server aggregates the contributions from these hospitals. Then the updated global model is shared with the local hospitals and they can continue local training. The most typical case is that a local hospital learns the medical experience of users from the diagnosis and treatment information of different users, providing relevant information on disease management for users according to the information with self-optimization. Vertical federation learning is suitable for situations where users in different domains have common data (data characteristics are not consistent). The training steps are as follows: 1) The coordinator is responsible for distributing the public key so that only coordinator can decrypt it; 2) participants carry out homomorphic encryption and interaction of the aligned samples, as well as calculate their respective gradient and loss values respectively; 3) after participants are calculated, they are sent to privacy-preserving entity (a mask or noise will be added at this time to avoid leaking); 4) privacy-preserving entity decrypts it, sending it back to participants which unmask and update their models. The main advantage of federated learning is that it establishes a global model without directly sharing datasets, preserving patient privacy across sites