Artificial intelligence supported clinical tool design for the diagnosis of Polycystic Ovary Syndrome

The aim of this study is to develop an artificial intelligence (AI)-assisted clinical decision support tool for the diagnosis of Polycystic Ovary Syndrome (PCOS), a significant endocrine disorder affecting women's reproductive health. In the literature, PCOS diagnosis is generally made using clinical symptoms, ultrasound imaging, and hormonal analyses. However, this process is time-consuming and largely depends on patient-physician interaction. In this thesis, a system has been designed that identifies missing information based on patients' text-based expressions and dynamically generates follow-up questions. BioBERT and SciBERT, two pre-trained language models specifically developed for the biomedical domain, were utilized. Unlike traditional large language models (LLMs), the proposed system estimates PCOS risk as a percentage based on patient data and supports decision-making by validating findings through scientific literature. The system can also suggest differential diagnoses for conditions such as NCCAH, hyperprolactinemia, and Cushing's syndrome, which present with symptoms similar to PCOS. To evaluate the quality of the system's generated explanations, natural language processing (NLP) metrics such as BLEU, ROUGE, and BERTScore were used. In conclusion, this study demonstrates the potential of AI-based interactive systems in PCOS diagnosis through NLP-driven dynamic querying and risk prediction, offering an innovative contribution to the literature.