This page was translated automatically from Turkish. Read the original

Bu sayfa, 26. Dünya Psikiyatri Kongresi'nde sunulan çalışmanın yapay zekâ desteğiyle hazırlanmış Türkçe özetidir. Çalışmanın sahibi sunan yazarlardır; atıf için orijinal kaynağa başvurun.

EV0820

Predicting dimensional psychiatric symptoms using eight-channel electroencephalography and an emotion manipulation paradigm

PREDICTING DIMENSIONAL PSYCHIATRIC SYMPTOMS USING AN EIGHT-CHANNEL ELECTROENCEPHALOGRAM AND EMOTION MANIPULATION PARADIGM

Eight-channel EEG provides discrimination comparable to, but as weak as, 26-channel recording while reducing resource use by approximately two-thirds; pilot measurements using emotional stimuli show lower frontal EEG reactivity and higher resting heart rate in the major depressive disorder group.

The aim is to evaluate the efficiency of eight-channel recording and emotion-evoked physiological reactivity, given the weak individual-level predictive performance of EEG in major depressive disorder (MDD) and the constraints that high-density recordings place on clinical use. The study comprises three phases: a historical cohort of 65 patients with MDD who underwent emotion manipulation in the first phase; an expanded archive of 246 patients with MDD and 32 healthy controls (HCs) in the second phase (eight channels were extracted from the same 26-channel recordings); and a pilot involving resting-state measurements and guided emotional interaction in 10 patients with MDD and 12 HCs using a separate device in the third phase; the historical and prospective cohorts were not pooled. Validation compared eight and 26 channels within the same recordings, using subject-level five-fold cross-validation, three random seeds, and 5,000 bootstrap iterations; the endpoints were classification based on item 3 of the HAMD-17 and regression of the total score. At the individual level, the eight-channel logistic model yielded an AUC of 0.566 (95% confidence interval, 0.515 to 0.656), while the 26-channel model yielded an AUC of 0.572 (95% confidence interval, 0.460 to 0.611); all 16 HAMD regression analyses produced negative R-squared values, and the deep learning models did not outperform the eight-channel logistic baseline model. In terms of efficiency, the reported electrode count was 26 versus 8 (a 69.2% reduction), estimated setup time was 39 versus 12 minutes (a 69.2% reduction), EEGNet FLOPs were 1.37e8 versus 4.30e7 (a 68.7% reduction), and EEGNet CPU time was 3.19 versus 1.99 ms (a 37.5% reduction). In the prospective pilot, findings during emotional interaction were d = -1.54, P = 0.030 for F4 alpha and d = -1.39, P = 0.040 for frontal theta; resting heart rate was 86.7 versus 71.5 beats/minute, d = +1.17, P = 0.027; and positive affect was 0.042 versus 0.127, d = -0.90, P = 0.053 (not significant). The authors position the method solely as an exploratory adjunctive tool and state that it cannot be used for standalone diagnosis or suicide risk prediction. Limitations include five unusable interaction EEG sessions and variation in the analyzed sample by modality, unadjusted pilot P values, medication effects and the absence of symptom scores, and the lack of evaluation of model transfer across devices.

Tüm WCP 2026 özetleri