Electrocardiogram (ECG) feature extraction is fundamental for detecting pathological conditions, yet traditional methods based on statistical summaries often fail to capture subtle morphological alterations. In this study, we adopt a functional spectral approach to analyze multichannel ECG data, preserving its temporal structure. We compare two methodologies for estimating the spectral density operator: Graphical Functional Principal Component Analysis (GFPCA), which computes cross-covariance at individual time points, and freqdom.fda, which extracts spectral features from basis expansion coefficients. In a case study of detachment episodes, freqdom.fda shows a better bias-variance tradeoff, effectively filtering noise while preserving key spectral patterns distinguishing regular and symptomatic states.