Paper deep dive
BandRouteNet: An Adaptive Band Routing Neural Network for EEG Artifact Removal
Phat Lam
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 99%
Last extracted: 6/21/2026, 8:25:41 AM
Summary
BandRouteNet is an adaptive, frequency-aware neural network designed for EEG artifact removal, specifically targeting EOG and EMG interference. The architecture utilizes a dual-path approach: a Band-specific Denoiser that performs band-wise denoising with an adaptive artifact routing mechanism, and a Full-band Conditioner that provides global temporal context and signal-level refinement via Feature-wise Linear Modulation (FiLM). Experimental results on the EEGDenoiseNet benchmark show that BandRouteNet outperforms several deep learning baselines in RRMSE and SNR improvement while maintaining high parameter efficiency (0.2M parameters).
Entities (9)
Relation Signals (6)
BandRouteNet → contains → Band-specific Denoiser
confidence 100% · The model contains two cooperative components: a Band-specific Denoiser, and a Full-band Conditioner.
BandRouteNet → contains → Full-band Conditioner
confidence 100% · The model contains two cooperative components: a Band-specific Denoiser, and a Full-band Conditioner.
EOG → contaminates → EEG
confidence 100% · Electroencephalography (EEG) is highly susceptible to artifact contamination, such as electrooculographic (EOG)
BandRouteNet → evaluatedon → EEGDenoiseNet
confidence 100% · Extensive experiments on the EEGDenoiseNet benchmark dataset demonstrate that BandRouteNet outperforms other methods
Band-specific Denoiser → uses → Artifact Router
confidence 100% · the Artifact Router and Band Denoiser share parameters across all bands.
Full-band Conditioner → modulates → Band-specific Denoiser
confidence 90% · producing both conditional parameters for modulating the band-wise pathway
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Electroencephalography (EEG) is highly susceptible to artifact contamination, such as electrooculographic (EOG) and electromyographic (EMG) interference, which severely degrades signal quality and hinders reliable interpretation in applications including neurological diagnosis, brain-computer interfaces (BCIs), etc. Effective EEG denoising remains challenging because different artifact sources exhibit diverse and temporally varying distributions, together with distinct spectral characteristics across frequency bands. To address these issues, we propose BandRouteNet, an adaptive frequency-aware neural network for EEG denoising that jointly exploits band-specific processing and full-band contextual modeling. The proposed model performs band-wise denoising to explicitly capture frequency-dependent artifact patterns. Within this framework, we introduce a routing mechanism that adaptively determines where and to what extent denoising should be applied across temporal locations within each frequency band. In parallel, a full-band conditioner directly processes the original noisy EEG to extract global temporal context, producing both conditional parameters for modulating the band-wise pathway and a coarse-grained signal-level refinement to supplement the final reconstruction. Extensive experiments on the EEGDenoiseNet benchmark dataset demonstrate that BandRouteNet outperforms other methods under EOG, EMG, and mixed-artifact conditions in terms of Relative Root Mean Square Error (RRMSE) and Signal-to-Noise Ratio Improvement (SNR$_{\text{imp}}$) under unified experimental settings, while remaining highly parameter-efficient with only 0.2M trainable parameters. These results highlight its strong potential for high-performance EEG artifact removal in resource-constrained applications.
Tags
Links
- Source: https://arxiv.org/abs/2604.24428v1
- Canonical: https://arxiv.org/abs/2604.24428v1
Trouble viewing inline? Open PDF directly →
Full Text
39,854 characters extracted from source content.
Expand or collapse full text
BandRouteNet: An Adaptive Band Routing Neural Network for EEG Artifact Removal Phat Lam P. Lam is with Ho Chi Minh City University of Technology, Vietnam. Abstract Electroencephalography (EEG) is highly susceptible to artifact contamination, such as electrooculographic (EOG) and electromyographic (EMG) interference, which severely degrades signal quality and hinders reliable interpretation in applications including neurological diagnosis, brain–computer interfaces (BCIs), etc. Effective EEG denoising remains challenging because different artifact sources exhibit diverse and temporally varying distributions, together with distinct spectral characteristics across frequency bands. To address these issues, we propose BandRouteNet, an adaptive frequency-aware neural network for EEG denoising that jointly exploits band-specific processing and full-band contextual modeling. The proposed model performs band-wise denoising to explicitly capture frequency-dependent artifact patterns. Within this framework, we introduce a routing mechanism that adaptively determines where and to what extent denoising should be applied across temporal locations within each frequency band. In parallel, a full-band conditioner directly processes the original noisy EEG to extract global temporal context, producing both conditional parameters for modulating the band-wise pathway and a coarse-grained signal-level refinement to supplement the final reconstruction. Extensive experiments on the EEGDenoiseNet benchmark dataset demonstrate that BandRouteNet outperforms other methods under EOG, EMG, and mixed-artifact conditions in terms of Relative Root Mean Square Error (RRMSE) and Signal-to-Noise Ratio Improvement (SNRimp_imp) under unified experimental settings, while remaining highly parameter-efficient with only 0.2M trainable parameters. These results highlight its strong potential for high-performance EEG artifact removal in resource-constrained applications. Keywords— EEG artifact removal, Electrooculography (EOG), Electromyography (EMG), EEGDenoiseNet, Routing Mechanism, Band-specific denoising, Full-band conditioner. I Introduction Electroencephalography (EEG) is a widely used non-invasive tool for monitoring brain activity due to its high temporal resolution, low cost, and portability. It supports a wide range of applications, including brain–computer interfaces (BCIs), neurological monitoring, sleep analysis, and clinical diagnosis [8]. However, EEG recordings are highly susceptible to contamination, particularly from electrooculographic (EOG) and electromyographic (EMG) artifacts, which can severely distort signal morphology, spectral content, and downstream interpretability [10]. A large body of work has focused on EEG denoising. Classical approaches mainly rely on explicit assumptions about artifact generation or signal structure. For example, Regression and adaptive filtering methods use reference channels, such as EOG, to estimate and subtract artifacts, but their performance depends strongly on the availability and quality of these references. [14]. Blind source separation methods particularly Independent Component Analysis (ICA), have been widely used for ocular and muscular artifact removal; however, they often require multichannel recordings, component selection heuristics [4, 7]. Time–frequency and decomposition-based approaches, such as Wavelet Transforms and Empirical Mode Decomposition (EMD), offer greater flexibility for non-stationary EEG signals, but they are sensitive to threshold selection, mode mixing, and decomposition instability [6, 5]. More recently, deep learning has emerged as a promising alternative for EEG denoising, as it can learn a direct mapping from contaminated EEG to clean EEG without heavily relying on hand-crafted priors. Zhang et al. [20] established end-to-end baselines using fully connected networks (FCNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Subsequent studies explored more advanced architectures, including the transformer-based EEGDNet [12], the embedding-separation framework DeepSeparator [18], the U-Net-based intepretable archiecture LRR-Unet [19], GAN-based approaches for artifact suppression [1, 17], etc. Although these methods have shown strong denoising performances, several limitations has remained, suggesting rooms for prospective improvements. First, frequency-aware modeling of EEG artifacts is still underexplored. Different artifact sources exhibit markedly different spectral characteristics (e.g. EOG are typically concentrated in low-frequency bands [2], whereas EMG are broader-band and often more prominent in higher frequencies [3]). Explicitly modeling such band-dependent patterns can therefore provide more targeted and effective denoising. Second, a large number of existing denoisers apply relatively uniform processing over the entire signal [13], which may under-correct severely contaminated regions or over-smooth clean segments. This motivates the need for an adaptive mechanism that can track artifact intensity over time while remaining aware of its frequency-specific behavior. Third, current deep models often emphasize either temporal sequence modeling or transformed-domain processing [13], but rarely integrate both in a manner tailored to EEG artifacts. Since EEG is inherently both temporally dynamic and spectrally structured, time-domain modeling capture temporal continuity but may overlook band-specific spectral structure, whereas frequency-centered approaches can isolate artifact-dominant bands but may weaken global temporal dependencies when bands are treated too independently. Motivated by these observations, we propose BandRouteNet, an adaptive EEG denoising network that combines frequency-aware band-wise processing with full-band temporal conditioning. The model contains two cooperative components: a Band-specific Denoiser, which operates on decomposed EEG bands to suppress artifacts in a frequency-aware manner, and a Full-band Conditioner, which processes the original full-band signal to provide global temporal guidance and cross-band contextual information. For band-wise denoising, instead of enforcing uniform denoising over all temporal regions, we further introduce an artifact routing mechanism that adaptively modulates the denoising strength according to the temporal and spectral characteristics of contamination. This design is intended to preserve cleaner neural structure while allocating stronger correction to severely corrupted band–time regions. The main contributions of this work are as followed: • First, we propose BandRouteNet, a frequency-aware and temporally adaptive denoising architecture that integrates band-specific processing with full-band contextual conditioning, enabling the exploration of both spectral structure and temporal dynamics for EEG artifact removal. • Second, we introduce an adaptive artifact routing mechanism that learns soft, time-varying denoising emphasis across latent band features, making the denoising process more selective, interpretable, and suitable for non-stationary artifact patterns. • Through extensive experiments on the EEGDenoiseNet benchmark, we show that the proposed method achieves superior denoising performance over other methods across EOG, EMG, and mixed-noise conditions under unified evaluation settings, while remaining highly parameter-efficient, supporting its potential for practical EEG denoising. I Method I-A The Overall Architecture (a) Overall framework of BandRouteNet. (b) Detailed structures of the main component blocks. Figure 1: Overview of the proposed BandRouteNet architecture for EEG denoising. The left subfigure illustrates the overall framework, while the right subfigure presents the detailed structures of the main component blocks. The model consists of two key modules: a Band-specific Denoiser, which removes artifacts within each decomposed EEG band, and a Full-band Conditioner, which processes the original full-band signal to capture global temporal dynamics and cross-band dependencies, thereby guiding and enhancing band-wise denoising. The proposed BandRouteNet is a dual-path EEG denoising framework that integrates band-wise decomposition with full-band contextual modeling. Given a noisy EEG segment ∈ℝB×1×TX ^B× 1× T, where B and T denote the batch size and segment length, respectively, the network estimates a clean reconstruction ^∈ℝB×1×T Y ^B× 1× T through two complementary components: a Band-specific Denoiser and a Full-band Conditioner. As shown at Figure 1a, the proposed BandRouteNet analyses the input EEG from two complementary views. First, the signal is decomposed into multiple frequency bands to explicitly capture band-dependent artifact characteristics. Second, the original full-band signal is modeled directly to exploit global temporal structure and inter-band context. Based on these two views, denoising is performed cooperatively: the band-specific pathway focuses on targeted artifact suppression within individual bands, while the full-band pathway provides global guidance to enhance and refine the band-wise reconstruction. The band-specific pathway follows an encode–route–reconstruct paradigm. Each decomposed band is first mapped into a latent representation, from which a band denoiser produces a candidate refined feature and an Artifact Router estimates the degree to which this refinement should be applied via a routing mechanism. Their interaction yields a routed latent representation that adaptively controls denoising strength across time and frequency bands. The routed band features are then passed through a cross-band fusion module to model dependencies among frequency bands. The refined band features are decoded and aggregated to form a band-wise reconstruction of the clean EEG. In parallel, the full-band pathway generates an coarse-grain signal-level refinement and extracts conditional parameters from the original noisy signal to modulate the band-wise features with global contextual information. The final denoised output is obtained by adaptively fusing the outputs of these two pathways. I-B Full-band Conditioner The Full-band Conditioner provides global temporal and cross-band context via adaptive feature modulation and signal-level refinement for the band-wise denoising pathway. This branch directly models the original noisy signal ∈ℝB×1×Tx ^B× 1× T and provides global guidance for more coherent denoising. First, the full-band noisy signal is encoded the input into a latent representation: f=ℰf(),f∈ℝB×C×T,h_f=E_f(x), _f ^B× C× T, (1) where ℰf(⋅)E_f(·) denotes the encoder. The latent features are then processed by a temporal enhancement block to capture long-range dependencies: f=f(f),f∈ℝB×C×T.z_f=T_f(h_f), _f ^B× C× T. (2) where f(.)=Conv1D(GRU(Norm(.)))T_f(.)=Conv1D (GRU(Norm(.)) ) (3) Here, fz_f serves as the global contextual representation of the noisy EEG. From fz_f, the branch produces three types of outputs. First, a decoder generates a coarse full-band refinement signal. This signal provides a signal-level correction complementary to the band-wise reconstruction: f=f(f),f∈ℝB×1×T.d_f=D_f(z_f), _f ^B× 1× T. (4) Second, a temporal gating head predicts =σ(ℋλ(f)),∈ℝB×1×T λ=σ(H_λ(z_f)), λ ^B× 1× T (5) where ℋλ(.)=Conv1D(GELU(Norm(Conv1D(.))))H_λ(.)=Conv1D (GELU(Norm(Conv1D(.))) ) (6) σ(⋅)σ(·) denotes the sigmoid activation. The gate λ controls how much the full-band refinement temporally contribute to the final denoised output at the final fusion stage. Third, a Conv1D-based projection head outputs feature modulation parameters: [,]=ℋproj(f),[ τ, ψ]=H_proj(z_f), (7) where ,∈ℝB×C×T τ, ψ ^B× C× T. These parameters are utilized to condition the band latent features at the band-specific pathway. I-C Band-specific Denoiser The Band-specific Denoiser exploits the spectral heterogeneity of EEG artifacts by processing each frequency band independently, rather than forcing a single shared operation to handle all frequencies uniformly in the time domain. DFT decomposition: First, the input noisy signal ∈ℝB×1×TX ^B× 1× T is decomposed into K frequency bands using an Discrete Fourier Transform (DFT) decomposition module. Specifically, for each signal ∈ℝTx ^T, the DFT is computed as x^[f]=∑t=0T−1x[t]e−j2πft/T,f=0,1,…,T−1. x[f]= _t=0^T-1x[t]e^-j2π ft/T, f=0,1,…,T-1. (8) Then, the spectrum is divided into K bands by band masks kk=1K\M_k\_k=1^K, and each band-specific signal is reconstructed by k=IDFT(k⊙^),k=1,2,…,K.x_k=IDFT (M_k x ), k=1,2,…,K. (9) Thus, the band-specific representation is written as b=1,2,…,K∈ℝB×K×T.X_b=\x_1,x_2,…,x_K\ ^B× K× T. (10) The Band-specific Denoiser processes each decomposed band independently in a shared latent space. For the k-th band, the individual band input k∈ℝB×1×Tx_k ^B× 1× T is mapped into a latent representation by a shared encoder: k=ℰb(k),k∈ℝB×C×Tu_k=E_b(x_k), _k ^B× C× T (11) Band Adapter: This module denoises the latent representation of each decomposed frequency band. Although one band adapter is applied to each band, the Artifact Router and Band Denoiser share parameters across all bands. Given the encoder features from k-th band ku_k, we add a learnable band identity embedding k∈ℝ1×C×1b_k ^1× C× 1 to explicitly distinguish it when different frequency bands in the shared latent space: ~k=k+k u_k=u_k+b_k (12) Feature-wise Linear Modulation To inject global context information from full-band signal into band-specific processing, we adopt Feature-wise Linear Modulation (FiLM) [11]. Given the latent representation of the k-th band, referred as ku_k, and the modulation parameters (,)( τ, ψ) produced by the Full-band Conditioner, the conditioned feature is computed as k=FiLM(~k,,)=~k⊙(1+tanh())+e_k=FiLM( u_k, τ, ψ)= u_k (1+ ( τ) )+ ψ (13) Through FiLM, the global are leveraged to adapt the scale and shift of each band-specific channel over time. The conditioned features are then passed to two parallel modules: an Artifact Router and a Band Denoiser. The Artifact Router This module estimates the denoising strength for each band-specific latent feature at every temporal location. Given the conditioned latent representation k∈ℝB×C×Te_k ^B× C× T, the router predicts a soft routing mask: k=ℛ(k),k∈[0,1]B×C×T.g_k=R(e_k), _k∈[0,1]^B× C× T. (14) To capture the temporal dynamics and context dependence of EEG artifacts, the Artifact Router jointly models local temporal context and global channel context. The local branch employs 1D depthwise-pointwise convolutions, while the global branch uses temporal average pooling followed by convolution layers. Their outputs are fused and passed through a sigmoid activation to produce the routing mask. This design aim to yield a time-varying, channel-aware control signal that determines where and to what extent denoising should be applied. The Band Denoiser This module is responsible for generating refined latent features for each decomposed band. Given the conditioned band latent k∈ℝB×C×Te_k ^B× C× T, the denoiser predicts k=ℬ(k),k∈ℝB×C×Tf_k=B(e_k), _k ^B× C× T (15) The module comprises a pre-convolution block, a GRU-based temporal layer, and a post-convolution block within a residual framework. For the k-th band, the Band Denoiser generates a candidate refinement feature kf_k rather than directly replacing the input representation. This design decouples refinement generation from refinement selection, enabling the network to learn artifact-suppressed features while allowing the Routing Mechanism to adaptively determine the extent to which kf_k is applied, attenuated, or suppressed. The Routing Mechanism The routing mechanism combines the original conditioned feature ke_k, the denoised proposal kf_k, and the routing mask kg_k to generate the final routed latent representation: k=(1−k)⊙k+k⊙k,z_k=(1-g_k) _k+g_k _k, (16) where ⊙ denotes element-wise multiplication. This formulation performs adaptive interpolation between the original feature and the denoised proposal: k→0g_k→ 0 preserves the original representation, whereas k→1g_k→ 1 emphasizes stronger denoising. Consequently, kg_k acts as a data-dependent control signal that regulates denoising strength for each band. Cross-band fusion: To model these inter-band dependencies, we employ a cross-band fusion module after band-wise routing. Given the stacked band features ∈ℝB×K×C×TZ ^B× K× C× T, the module refines them using two cascaded operations including a temporal mixing layer and a band mixing layer. First, temporal mixing is applied independently within each band to model intra-band temporal dynamics: (t)=ℳtemp().Z^(t)=M_temp(Z). (17) Next, in the band mixing stage, features at each time step interact across the K bands through multi-head self-attention (MHSA) block, followed by a position-wise feed-forward network: ′=ℳband((t)).Z =M_band(Z^(t)). (18) For each time step, the K band features are processed by multi-head self-attention block (MHSA) [16]. Specifically, for the m-th head, Headm=Softmax(mm⊤dh)m,Head_m=Softmax ( Q_mK_m d_h )V_m, (19) and the outputs of all heads are concatenated and projected to obtain the final representation of the block. I-D Final fusion After cross-band fusion, each band feature is decoded into a denoised band signal: ^k=b(k′),^k∈ℝB×1×T y_k=D_b(z_k ), y_k ^B× 1× T (20) and all decoded bands are concatenated to form ^bands∈ℝB×K×T Y_bands ^B× K× T (21) The aggregated band-wise denoised signal ^band Y_band is obtained by summing over K bands. The final denoised signal is the fusion that combines fine-grained band-wise denoising and coarse-grained global denoised signal from the Full-band conditioner as: ^=^band+⊙f Y= Y_band+ λ _f (22) where the condition parameter λ control the contribution of globally-denoised signal to the final denoised signal. I Experimental Settings I-A Datasets To evaluate the denoising performance of the proposed model, we adopt the EEGDenoiseNet benchmark [20], which is widely used in deep learning-based EEG denoising studies. The dataset consists of 4515 clean EEG signals, 3400 electrooculogram (EOG) artifact signals, and 5598 electromyogram (EMG) artifact signals. Each sample has a duration of 2 seconds with a sampling rate of 256 Hz. Let x denote the clean EEG signal and N denote the artifact signal (EOG or EMG). A contaminated EEG signal y is generated as: y=x+λN,y=x+λ N, (23) where λ controls the noise intensity. The signal-to-noise ratio (SNR) is adjusted by changing the parameter λ as following: SNR=10logRMS(x)RMS(λ⋅n)SNR=10 RMS(x)RMS(λ· n) (24) in which the Root Mean Squared (RMS) value is defined as RMS(g)=1N∑i=1Ngi2RMS(g)= 1N _i=1^Ng_i^2 (25) TABLE I: Performance Comparison of Average Performances across All SNR Levels. The smaller RMSSEtRMSSE_t, RMSSEsRMSSE_s, and the Larger CC, SNRimpSNR_imp the Better Denoising Effect. Best Results are in Bold; Second-best Ones are Underlined Method Parameter (M) EOG dataset EMG dataset Mixed EOG/EMG dataset RRMSEt_t RRMSEs_s C SNRimp_imp RRMSEt_t RRMSEs_s C SNRimp_imp RRMSEt_t RRMSEs_s C SNRimp_imp FCNN [20] 1.051 0.5570 0.5865 0.8111 10.8011 0.6176 0.6687 0.7849 9.6891 0.6887 0.7362 0.7123 11.7388 Simple-CNN [20] 16.82 0.4437 0.4432 0.8847 12.6351 0.7215 0.7263 0.7255 8.4879 0.7787 0.7258 0.6763 10.7515 1D-ResCNN [15] 8.46 0.4184 0.4046 0.8966 13.1761 0.7510 0.7093 0.7076 8.2550 0.8674 0.9065 0.6625 10.0200 RNN-LSTM [20] 0.788 0.6903 0.7231 0.7176 8.6099 0.6758 0.7215 0.7384 8.8269 0.7522 0.8123 0.6644 10.8124 EEGDnet [12] 0.895 0.4406 0.4125 0.8907 12.4712 0.6554 0.6196 0.7456 9.0505 0.6897 0.6210 0.7125 11.5704 Deep Separator [18] 0.032 0.4944 0.5484 0.8658 11.3920 0.6982 0.6899 0.7396 8.6998 0.7593 0.7174 0.6814 10.7386 LRR-UNet [19] 3.19 0.4322 0.4311 0.8967 12.7577 0.6415 0.5827 0.7564 9.3746 0.7144 0.6328 0.7068 11.3037 BandRouteNet (Ours) 0.20 0.3831 0.3797 0.9156 13.9819 0.5962 0.5276 0.7802 10.0041 0.6605 0.5998 0.7359 11.9310 TABLE I: Ablation Study Results on EOG Dataset Experiments RRMSEt_t RRMSEs_s C SNRimp_imp Full model 0.3831 0.3797 0.9156 13.9819 W/o fullband conditioner 0.3907 0.3979 0.9106 13.2324 W/o artifact routing (=g=1) 0.4091 0.4013 0.9112 13.7610 W/o cross-band fusion 0.3956 0.4009 0.9139 13.8310 W/o band identity embedding 0.3867 0.3998 0.9130 13.8392 From the samples belong to each types of signal, three types of dataset are constructed: EOG Dataset The EOG-contaminated dataset is created from 3400 EEG segments and 3400 EOG segments. At specified signal-to-noise ratio (SNR) levels, pure EEG signals were linearly combined with ocular artifacts to generate EOG-contaminated signals. EMG Dataset In the EMG setting, EEG signals are randomly reused to match the number of EMG segments, resulting in 5598 samples from the EOG-contaminated EEG signals. Similarly, using specified SNR levels, pure EEG signals were linearly combined with EMG artifacts to generate EMG artifact-contaminated signals. Mixed EOG/EMG Dataset For the mixed-artifact setting, EEG, EOG, and EMG signals are all aligned to a common size equal to the maximum number of artifact samples. For each SNR level, both EOG/EMG are scaled independently to that level and linearly combined: y=x+λ(NEOG+NEMG)y=x+λ(N_EOG+N_EMG) (26) Following the dataset construction protocol described in the official benchmark study [20], each EEG-artifact pair is augmented across 10 signal-to-noise ratio (SNR) levels ranging from −7-7 to 22 dB to enhance training diversity. The resulting datasets are then divided into training, validation, and test sets with a ratio of 8:1:1. Furthermore, both clean and noisy signals are standardized using the standard deviation of the corresponding noisy signal, as defined below: x^=xσy,y^=yσy x= x _y, y= y _y (27) I-B Evaluation Metrics To quantitatively assess the denoising performance of the proposed model, we adopt the standard evaluation metrics, including the Temporal Root Mean Squared Error (RRMSEt), Spectral Root Mean Squared Error (RRMSEs), and Correlation Coefficient (C). In addition, we also report the Signal-to-noise Ratio Improvement (SNRimp) to measure the gain in signal quality before and after denoising. Temporal Relative Root Mean Squared Error RRMSEt evaluates the waveform-level reconstruction error between the denoised signal y y and the clean reference signal y: RRMSEt=RMS(y^−y)RMS(y)RRMSE_t= RMS( y-y)RMS(y) (28) Spectral Relative Root Mean Squared Error RRMSEs measures the discrepancy between the spectra of the denoised and clean signals: RRMSEs=RMS(S(y^)−S(y))RMS(S(y))RRMSE_s= RMS\! (S( y)-S(y) )RMS\! (S(y) ) (29) where S(⋅)S(·) denotes the Power Spectral Density (PSD) of the signal. Correlation Coefficient The correlation coefficient (C) measures the linear similarity between the denoised signal and the clean target: C=∑i=1N(yi−y¯)(y^i−y^¯)∑i=1N(yi−y¯)2∑i=1N(y^i−y^¯)2C= _i=1^N (y_i- y ) ( y_i- y ) _i=1^N (y_i- y )^2 _i=1^N ( y_i- y )^2 (30) where y¯ y and y^¯ y denote the mean values of y and y y, respectively. SNR improvement To further evaluate denoising effectiveness, we compute the SNR improvement (SNR imp), defined as the difference between the output SNR and the input SNR: SNRimp=SNRout−SNRinSNR_imp=SNR_out-SNR_in (31) Given the clean signal x, noisy observation y, and denoised output y y, the input and output SNR values are computed as SNRin _in =10log10(P(x)P(y−x)), =0 _10 ( P(x)P(y-x) ), (32) SNRout _out =10log10(P(x)P(y^−x)). =0 _10 ( P(x)P( y-x) ). where P(⋅)P(·) denotes the average signal power. I-C Implementation Details The proposed model was implemented in PyTorch and trained on an NVIDIA Tesla T4 GPU. We used the AdamW optimizer [9] with an initial learning rate of 1×10−31× 10^-3 and a weight decay of 1×10−41× 10^-4. The model was trained for 15 epochs using Mean Squared Error (MSE) as the loss function. For model configuration, the input noisy EEG signal was decomposed into K=6K=6 frequency bands: Delta (δ, 0–-4 Hz), Theta (θ, 4–-8 Hz), Alpha (α, 8–-12 Hz), Beta (β, 13–-30 Hz), Gamma (γ, 30–-80 Hz), and a higher band ϵε (80–-128 Hz). The hidden feature dimension was set to C=64C=64. Both the Band-specific Denoiser and the Full-band Conditioner adopt an Inception-based encoder–-decoder architecture to capture multi-scale features, as shown at the Figure 1b. Each module consists of two encoder stages and two decoder stages, with each stage incorporating two Inception1D blocks. IV Results and Discussion Figure 2: Denoising performance comparison on the EOG-contaminated segment Figure 3: Denoising performance comparison on the EMG-contaminated segment Figure 4: Denoising performance comparison on the mixed-noise segment To assess the effectiveness of the proposed BandRouteNet, we compare it against several neural network-based EEG denoising methods, including the benchmark baselines FCNN [20], SimpleCNN [20], RNN-LSTM [20], and 1D-ResCNN [20], as well as recent state-of-the-art architectures such as EEGDNet [12], Deep Separator [18], and LRR-Unet [19]. To ensure a fair comparison, all competing methods are reproduced and evaluated under a unified experimental protocol, including the same training, validation, and test splits, identical evaluation and visualization settings. (a) RMSSE temporal (b) RMSSE spectral (c) Correlation coefficient (d) SNR improvement Figure 5: Performance Metrics on different SNR Levels (EOG Dataset) (a) RMSSE temporal (b) RMSSE spectral (c) Correlation coefficient (d) SNR improvement Figure 6: Performance Metrics on different SNR Levels (EMG Dataset) (a) RMSSE temporal (b) RMSSE spectral (c) Correlation coefficient (d) SNR improvement Figure 7: Performance Metrics on different SNR Levels (Mixed EOG/EEG Dataset) (a) EOG-contaminated EEG signal (b) EMG-contaminated EEG signal (c) EOG/EMG-contaminated EEG signal Figure 8: Artifact routing visualization results under three noise conditions. In each sub-figure, the upper panels present the denoising results, while the lower panel shows the outputs of the Artifact Router module as a heatmap over time and frequency bands. IV-A Qualitative Evaluation in the Time and Frequency Domains We first conduct visual analysis to compare denoising performances of all methods under EOG, EMG, and mixed-noise conditions using reconstructed waveforms and PSD curves, as shown in Fig. 2, Fig. 3, and Fig. 4. Across all scenarios, BandRouteNet produces reconstructions that more closely follow the clean target (depicted in green) while effectively suppressing artifacts without introducing residual noise or over-smoothing. Its PSD curves also better match the clean spectra (depicted in green), indicating stronger preservation of EEG spectral characteristics. These results suggest that BandRouteNet achieves a favorable balance between artifact removal and intrinsic EEG signal preservation. IV-B Quantitative Comparison Across Evaluation Metrics We conduct quantitative experiments to compare the denoising performance of the proposed BandRouteNet against all competing methods. Table I reports the average results across all SNR levels on the three benchmark settings, namely the EOG, EMG, and mixed EOG/EMG datasets. The evaluation follows four standard metrics, including RMSSEtRMSSE_t, RMSSEsRMSSE_s, CC, and SNRimpSNR_imp. As shown in Table I, the proposed model achieves the best overall performance across all three datasets. On the EOG dataset, BandRouteNet attains the lowest RMSSEtRMSSE_t (0.3831) and RMSSEsRMSSE_s (0.3797), as well as the highest CC (0.9156) and SNRimpSNR_imp (13.9819), outperforming all competing methods. A similar trend is observed on the EMG dataset, where the proposed model again yields the best RMSSEtRMSSE_t (0.5962), RMSSEsRMSSE_s (0.5276), and SNR improvement (10.0041), while remaining highly competitive in CC (0.7802). On the mixed EOG/EMG dataset, which is the most challenging setting due to the coexistence of multiple artifact types, the proposed system continues to deliver the strongest overall performance, achieving the best RMSSEtRMSSE_t (0.6605), RMSSEsRMSSE_s (0.5998), CC (0.7359), and SNRimpSNR_imp (11.9310). These results demonstrate that the proposed architecture generalizes well across different contamination types and provides a consistently better trade-off between waveform fidelity, spectral preservation, correlation with the clean signal, and noise suppression. We further analyze the performance across different SNR levels, as shown in Fig. 5, Fig. 6, and Fig. 7. In general, all methods improve when the input SNR increases. Specifically, BandRouteNet maintains superior or highly competitive performance throughout the full SNR range. In particular, the proposed model consistently achieves lower temporal and spectral reconstruction errors, while preserving higher or competitive correlation and SNR improvement over most SNR levels. This advantage is especially evident in the low-SNR regime, where the denoising task is most challenging. As the SNR increases, the proposed model continues to exhibit stable and favorable performance, further demonstrating its generalization capability across different noise intensities. In addition to the strong denoising performance, BandRouteNet is also highly lightweight. The proposed model contains only 0.20 million trainable parameters, which is far smaller than several comparing methods, yet it still delivers clearly superior performance. Overall, these quantitative results show that BandRouteNet is not only more accurate than other methods, but also more parameter-efficient, showing potential to be a practical solution for high-performance EEG denoising under resource-constrained settings. IV-C Ablation Analysis of the Proposed Network Components To assess the contribution of each module to the model performance, we perform an ablation study on the EOG dataset by removing one component at a time from the full model, including the full-band conditioner, artifact routing gate, cross-band fusion module, and band identity embedding. The results reported in Table I show that each proposed component contributes to the final denoising quality. While the full model achieves the best performance on all metrics, with RMSSEtRMSSE_t of 0.3831, RMSSEsRMSSE_s of 0.3797, CC of 0.9156, and an SNRimpSNR_imp of 13.9819, all ablated variants show degraded performance. Notably, removing the artifact routing gate leads to the largest drop in performance (increases of 0.026 and 0.0216 in RMSSEtRMSSE_t and RMSSEsRMSSE_s, respectively, indicating that adaptive temporal routing is a key factor in the proposed architecture. Removing the full-band conditioner also causes a clear reduction in performance (with RMSSEtRMSSE_t and RMSSEsRMSSE_s increasing by 0.0076 and 0.0182, respectively), which verifies the importance of incorporating global temporal and cross-band contextual information in addition to band-wise processing. Other components such as cross-band fusion and band identity embedding positively contribute to the better reconstruction of artifact-free EEG signals. IV-D Visualization of the Artifact Routing Mechanism To analyze the proposed routing mechanism interpretably, we visualize the Artifact Router outputs under EOG, EMG, and mixed EOG/EMG noise conditions in Fig. 8. In each subfigure, the upper panel shows the denoising result, while the lower panel presents the corresponding router activation map. Larger heatmap values indicate stronger denoising emphasis on the corresponding band-specific features. The activations vary adaptively across both time and frequency. For EOG noise, strong responses mainly appear in low-frequency bands, especially Delta (δ, 0–4 Hz) and Theta (θ, 4–8 Hz), during intervals with prominent ocular artifacts. For EMG noise, the responses shift toward higher-frequency bands, consistent with the broadband, high-frequency nature of muscle artifacts. Under mixed noise, the routing map combines both patterns, showing that the model can respond to coexisting low- and high-frequency contamination. The activations are also temporally localized and smooth, suggesting that the router increases denoising mainly in corrupted regions while preserving cleaner signal segments. V Conclusion This paper proposed BandRouteNet, an adaptive neural network for EEG denoising. By combining band-specific denoising with routing-based adaptive refinement, complemented by full-band contextual conditioning, the proposed model effectively suppresses diverse artifacts and preserve the intrinsic structure of EEG signals. Extensive experiments under various noise conditions show that BandRouteNet achieves superior performance across various standard evaluation metrics, while remaining highly parameter-efficient. References [1] E. Brophy et al. (2022) Denoising eeg signals for real-world bci applications using gans. Volume 2 - 2021. External Links: Document, ISSN 2673-6195 Cited by: §I. [2] R. J. Croft et al. (2000) Removal of ocular artifact from the eeg: a review. 30 (1), p. 5–19. Cited by: §I. [3] I. I. Goncharova et al. (2003) EMG contamination of eeg: spectral and topographical characteristics. 114 (9), p. 1580–1593. Cited by: §I. [4] S. Halder, M. Bensch, J. Mellinger, M. Bogdan, A. Kübler, N. Birbaumer, and W. Rosenstiel (2007) Online artifact removal for brain-computer interfaces using support vector machines and blind source separation. 2007 (1), p. 082069. Cited by: §I. [5] Q. Hu et al. (2022) Single-channel eeg signal extraction based on dwt, ceemdan, and ica method. 16, p. 1010760. Cited by: §I. [6] C. Kaur, P. Singh, and S. Sahni (2021) EEG artifact removal system for depression using a hybrid denoising approach. 12 (4), p. 465. Cited by: §I. [7] Y. Li, P. T. Wang, M. P. Vaidya, R. D. Flint, C. Y. Liu, M. W. Slutzky, and A. H. Do (2021) Electromyogram (emg) removal by adding sources of emg (erase)—a novel ica-based algorithm for removing myoelectric artifacts from eeg. 14, p. 597941. Cited by: §I. [8] X. Liu, W. Wang, M. Liu, M. Chen, T. Pereira, D. Y. Doda, Y. Ke, S. Wang, D. Wen, X. Tong, et al. (2025) Recent applications of eeg-based brain-computer-interface in the medical field. 12 (1), p. 14. Cited by: §I. [9] I. Loshchilov and F. Hutter (2019) Decoupled weight decay regularization. External Links: 1711.05101 Cited by: §I-C. [10] M. M. N. Mannan, M. A. Kamran, and M. Y. Jeong (2018) Identification and removal of physiological artifacts from electroencephalogram signals: a review. 6, p. 30630–30652. Cited by: §I. [11] E. Perez et al. (2018) Film: visual reasoning with a general conditioning layer. In Proceedings of the AAAI conference on artificial intelligence, Vol. 32. Cited by: §I-C. [12] X. Pu et al. (2022) EEGDnet: fusing non-local and local self-similarity for eeg signal denoising with transformer. 151, p. 106248. External Links: ISSN 0010-4825 Cited by: §I, TABLE I, §IV. [13] V. A. Raj et al. (2025) A comprehensive review of deep learning models for denoising eeg signals: challenges, advances, and future directions. 7 (11), p. 1268. Cited by: §I. [14] S. Romero et al. (2009) Ocular reduction in eeg signals based on adaptive filtering, regression and blind source separation. 37 (1), p. 176–191. Cited by: §I. [15] W. Sun et al. (2020) A novel end-to-end 1d-rescnn model to remove artifact from eeg signals. 404, p. 108–121. External Links: ISSN 0925-2312 Cited by: TABLE I. [16] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin (2023) Attention is all you need. External Links: 1706.03762 Cited by: §I-C. [17] H. Wang et al. (2024) EEG signal denoising using pix2pix gan: enhancing neurological data analysis. Cited by: §I. [18] J. Yu et al. (2022-04) Embedding decomposition for artifacts removal in eeg signals. 19 (2), p. 026052. Cited by: §I, TABLE I, §IV. [19] X. Yue et al. (2025) LRR-unet: a deep unfolding network with low-rank recovery for eeg signal denoising. 31 (10), p. e70632. Cited by: §I, TABLE I, §IV. [20] H. Zhang et al. (2021) EEGdenoiseNet: a benchmark dataset for end-to-end deep learning solutions of eeg denoising. External Links: 2009.11662 Cited by: §I, §I-A, §I-A, TABLE I, TABLE I, TABLE I, §IV.