Data Augmentation and Classification of Functional Near-Infrared Spectroscopy Data Using Denoising Diffusion Probabilistic Model Fonksiyonel Yakin Kizilötesi Spektroskopisi Verilerinin Gürültü Giderici Difüzyon Olasiliksal Modeli ?Ile Artirimi ve Siniflandirilmasi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Turkey, 7 - 10 July 2026, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/siu71813.2026.11636910
- City: İstanbul
- Country: Turkey
- Keywords: brain-computer interface, data augmentation, diffusion model, functional near-infrared spectroscopy, synthetic data generation
- Middle East Technical University Affiliated: Yes
Abstract
This study proposes a denoising diffusion probabilistic model (DDPM) based method for synthetic generation of functional near-infrared spectroscopy (fNIRS) data. Yes - No stimulus paradigm data collected from 28 participants were processed through a comprehensive signal processing pipeline and synthetic epoch data were generated using a conditional one-dimensional U-Net architecture. With the proposed pipeline, features showed no statistically significant difference from real data. The findings from the classification demonstrate that DDPM based data augmentation yields statistically significant improvements in the performance of deep learning models.