A concept-enhanced, knowledge graph-guided framework for interpretable PCOS prediction: a case study in explainable biomedical artificial intelligence

View at publisher Summarise with VIRAT
Authors: Sudeepti Kulshrestha, Neeraj Kumar, Pulkit Verma, Kriti Sikri, Mayank Gangwar, Abhishek Sengupta, Suruchi Aggarwal, Priyanka Narad
Cited by: 0
References: 27

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

CONTEXT: Overcoming the disconnect between efficient Machine Learning (ML) techniques and its therapeutic applicability necessitates frameworks that provide clear, contextualized evidence. AIMS: The present work examines the efficacy of a conceptually-enhanced framework that combines biological domain expertise, knowledge graphs, and probabilistic inference with ML predictions. We employed this innovative strategy for predicting polycystic ovary syndrome (PCOS). METHODS: PCOS clinical data (541 patients) was sourced from Kaggle. Clinically relevant conceptual characteristics were developed and integrated with raw data using a knowledge graph (KG) employing TransE-based semantic proximity scoring and a Bayesian network (BN) to characterize conditional dependencies. Knowledge-driven concepts were included in subsequent ML procedures, including LazyPredict for model screening, and ensemble model construction from top 10 models based on accuracy using soft-voting ensemble classifier. Patient-specific, guideline-compliant explanations were designed to accompany all model predictions for improved interpretability. KEY RESULTS: The approach demonstrated robust predictive abilities, with the concept-integrated ensemble model demonstrating the strongest performance, with a five-fold stratified cross-validation accuracy of 93.65% (95% CI: 91.59-95.63%) and a ROC-AUC of 0.98 (95% CI: 0.96-1.00%). In addition to predicting accurately, the incorporation of features obtained from KG and BN significantly improved interpretability, with causal probabilities based on BN proving to be the most important feature. The framework facilitated patient-specific, concept-based interpretations consistent with clinical guidelines. CONCLUSION: The integration of biomedically-relevant concepts improves the interpretability of PCOS predictions while maintaining predictive performance. IMPLICATIONS: Our work proposes the use of clinician-aligned PCOS screening tools that provide transparent risk classification and informed treatment decisions.

Fields of Study & Subjects

Explainable Artificial Intelligence (XAI) Artificial Intelligence in Healthcare Ovarian function and disorders

Publication details

Journal / source
Reproduction Fertility and Development
Volume
38
Issue
5
Publication date
26 March 2026
Language
EN