Generating Synthetic Brain Signals of Parkinson's Disease Using a Conditional Diffusion-Based Approach Kosullu Difüzyon Tabanli Bir Yaklasim ile Parkinson Hastalarinin Beyin Sinyallerinin Sentetik Üretimi

Ekiz M., ÖZKURT T. E.

34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu71813.2026.11636699
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: conditional generative diffusion model, local field potentials, Parkinson's disease, synthetic brain signal generation, time-frequency analysis
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

Subcortical local field potentials (LFPs) are signals that play an important role in investigating pathological neural oscillations associated with Parkinson's disease. In particular, beta band activity in the 13-35 Hz range is considered one of the main electrophysiological biomarkers of the disease. However, the invasive surgical procedures required for LFP recordings and the limited accessibility of clinical data restrict data-driven machine learning studies. In this study, a conditional diffusion-based generative framework is proposed for synthesizing LFP signals of Parkinson's disease patients. Raw signals were transformed into time-frequency representations using the Short-Time Fourier Transform, and magnitude and phase components were used as model inputs. The model was conditioned on clinical variables including medication state, dominant hemisphere, akinetic-rigid (AR) score, and tremor score. The proposed method was evaluated using an open-access dataset containing recordings from 18 Parkinson's disease patients. Results show significant correlations between beta band power and clinical scores in the generated signals (Spearman ρ = 0.43-0.57). Furthermore, the Wilcoxon test confirmed that dopaminergic medication suppresses beta-band activity (p < 0.01). Pearson similarity analysis indicated moderate similarity (~0.5-0.6) between synthetic and real signals.