سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Natural Language Processing and Speech Processing Integration: Toward A Point-of-Care Framework for Early Detection of Alzheimer’s Disease

نویسندگان :

Aslan Modir¹ Fatemeh Shalchizadeh² Armin Ghasimi³ Sina Shamekhi⁴

1- دانشگاه صنعتی تبریز 2- دانشگاه صنعتی تبریز 3- دانشگاه صنعتی تبریز 4- دانشگاه صنعتی تبریز

کلمات کلیدی :

Alzheimer’s disease،Speech processing،Large Language Model،Machine Learning،Mild Cognitive Impairment،Natural Language Processing

چکیده :

Alzheimer’s disease (AD) is a major global health concern, with no definitive cure currently available. Recent medical advances have introduced pharmacological interventions that may slow neurodegenerative progression, particularly in the stage of mild cognitive impairment (MCI). However, existing biomarkers require specialized clinical facilities, making largescale MCI screening challenging. To address this limitation, we propose a fully automatic framework for MCI detection that integrates linguistic and acoustic features extracted from spontaneous picture description tasks. A Large Language Model (LLM) has been used to analyze the semantic and syntactic structure of speech, capturing representations of semantic, episodic, and working memory, as well as emotional cues affected by AD. Simultaneously, acoustic processing has been adopted to quantify changes in voice quality associated with neurodegeneration. The fused feature set is evaluated using three well-established classifiers. Validation in the TAUKADIAL Challenge dataset demonstrates that the proposed framework achieves an accuracy of 0.8 under a Leave-One Subject-Out cross-validation approach, representing the highest reported performance on the test set. Importantly, the framework relies on a single picture description task, utilizes a compact LLM, and is built entirely on open-source packages, making the framework user-friendly and suitable for real-world point-of-care applications.