Cognitive Monitoring in Postmenopausal Women Using AI- Based Speech Analysis: A Novel Approach to Early Dementia Screening

Authors

1 National Association of Iranian Gynecologists & Obstetricians (NAIGO), Mashhad, Iran.

2 SENSOMATT Lda., Castelo Branco, Portugal.

3 SENSOMATT Lda., Castelo Branco, Portugal .

10.22038/ijogi.2026.27614

Abstract

Introduction: Postmenopausal women are at increased risk of cognitive decline due to hormonal changes affecting memory and executive function. Traditional screening methods are limited by accessibility and scalability. This study presents a novel, contactless approach for early dementia detection through speech analysis using advanced AI models, offering a scalable solution for continuous cognitive monitoring.
Methods: We designed an AI-based call system that conducts scheduled telephone conversations with patients. Healthcare providers can register patients through a secure web platform to receive daily, weekly, or monthly automated voice interactions. These semi-structured conversations prompt patients to describe daily routines, recall memories, and reason through simple tasks. Calls are recorded and transcribed using Google Speech-to-Text API V2, chosen for its high accuracy in Persian. From the transcriptions, acoustic (pause duration, speech rate, pitch variation) and linguistic (semantic coherence, lexical diversity, pronoun use) features are extracted. The core analytical engine is based on GPT-4o by OpenAI, a state-of-the-art large language model capable of advanced contextual reasoning. GPT-4o evaluates the semantic relevance of answers, narrative clarity, and changes over time. A personalized baseline model monitors deviations in cognitive-linguistic behavior using historical data for each patient. Clinicians access patient summaries, risk alerts, and longitudinal trends via a secure dashboard.
Results: In a pilot with 30 Persian-speaking postmenopausal women aged 55–70, 26% demonstrated patterns associated with mild cognitive decline. Key indicators included prolonged speech pauses, reduced narrative structure, and semantic inconsistencies. GPT-4o was effective in detecting subtle linguistic shifts that aligned with early cognitive risk markers. The system achieved high engagement and acceptance, with over 90% of participants completing calls without support. Compared to traditional tools (e.g., MMSE), this method offered continuous, naturalistic assessment with no need for clinic visits or hardware installations.
Conclusion: This speech-based AI monitoring system offers an innovative and scalable method for early detection of cognitive decline in postmenopausal women. By combining accessible telephonic interaction with advanced LLM processing (GPT-4o), it enables low-barrier cognitive tracking suitable for widespread clinical deployment. A dedicated call center has been established, allowing gynecologists and clinics to enroll patients for automated follow-up. Future expansions will include multimodal analysis and multilingual support to enhance coverage and personalization.

Keywords

The Iranian Journal of Obstetrics, Gynecology and Infertility
Volume 28, Issue 11
January 2026
Pages 8-11

Files

Share

How to cite

Statistics