Detection of Alzheimer's Disease by Using Time-Frequency Representations of EEG Signals with Deep Learning
11th International Conference on Control Decision and Information Technologies-CODIT, Split, Hırvatistan, 15 - 18 Temmuz 2025, ss.195-199, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/codit66093.2025.11321577
- Basıldığı Şehir: Split
- Basıldığı Ülke: Hırvatistan
- Sayfa Sayıları: ss.195-199
- İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet
Özet
Alzheimer's disease (AD) is a neurodegenerative disorder and the most common type of dementia. It leads to impairments in cognitive functions and seriously affects quality of life. Early diagnosis of the disease is crucial for effective treatment and management. This study proposes a new method using a modified ResNet18 CNN architecture to detect and monitor AD using electroencephalography (EEG) signals. The standard ResNet18 CNN architecture was simplified to use fewer layers and lower filter degrees to expedite the training procedures. In the proposed method, the scalogram images obtained using the Continuous Wavelet Transform (CWT) from 5 sec EEG segments of the AD and control groups are used as input to the modified ResNet18 CNN architecture. 2D time-frequency images of EEG segments are generated using both Bump wavelet CWT and the Short-Time Fourier Transform (STFT), for comparison. Calculated images are used to train the standard ResNet18, and modified ResNet18 CNN architectures to classify the EEG segments. Experimental results show that the CWT approach achieved higher performance compared to the STFT, and the proposed modified ResNet18 CNN architecture (93.74% accuracy) demonstrated more balanced performance than other architectures, exhibiting no overfitting, and completed the training much faster than other models, providing significant time savings.