Phase 4 · Signal Conditioning & Clinical Robustness
Artifact Mitigation, Motion Correction & SNR Optimization
Decontaminating physiological recordings from the messy reality of the clinic: Rigid-body prospective motion correction, independent component analysis (ICA), adaptive filtering, and motion tolerance in un-sedated pediatric neuroimaging.
Course: IDNE 701 (Fall 2026)
Date: Oct 30, 2026 (Week 10)
Duration: 80 Minutes
Instructor: Mark Bolding, PhD
01. Sources of Real-World Signal Corruption
In textbook equations, neural signals are assumed to be corrupted only by zero-mean white Gaussian noise $\mathcal{N}(0, \sigma^2)$. In live clinical recordings, real-world signals are engulfed by physiological rhythms, gross patient movement, and environmental interference:
🫀 Physiological Pulsatility
Cardiac pulsatility ($1 - 2.5\,\text{Hz}$ in neonates) drives rhythmic brain expansion and acoustic tissue Doppler shifting. Respiratory chest motion ($0.3 - 0.7\,\text{Hz}$) causes periodic magnetic field modulation ($\Delta B_0$) and venous pooling.
⚡ Gross Patient Movement
Head rotation and swallowing produce non-rigid shear displacement between skin and sensors, leading to catastrophic phase errors in k-space and decoupling baseline spikes in optical optodes.
🔌 Environmental Interference
$60\,\text{Hz}$ powerline harmonics, incubator switching power supplies, infusion pump motors, and blue bilirubin phototherapy lamps.
02. Mathematical Models of Rigid-Body Motion Correction
In 3D spatial neuroimaging, subject movement is modeled as a 6-degree-of-freedom (6-DOF) affine rigid-body coordinate transformation:
$$\vec{r}' = \mathbf{R}(\theta_x, \theta_y, \theta_z) \vec{r} + \vec{T}(t_x, t_y, t_z)$$
Equation 1: 3D Rigid-body spatial transformation
In k-space, translation introduces a linear phase ramp, whereas rotation rotates the k-space coordinate grid itself:
$$\mathcal{F}\{\rho(\vec{r} - \vec{T})\} = S(\vec{k}) \cdot e^{-i 2\pi \vec{k} \cdot \vec{T}}$$
Equation 2: Fourier Shift Theorem in spatial frequency domain
Prospective Motion Correction (PROMO): High-speed optical tracking cameras or orbital k-space navigators measure $\mathbf{R}$ and $\vec{T}$ in real time ($< 10\,\text{ms}$), dynamically updating the RF excitation center frequency and gradient coordinate axes before each readout.
03. Blind Source Separation: ICA & Adaptive Filtering
In multi-channel electrophysiology and optical imaging, recorded signals $\mathbf{x}(t)$ represent a linear mixture of true cortical neural generators $\mathbf{s}_{\text{neural}}(t)$ and non-neural artifact generators $\mathbf{s}_{\text{artifact}}(t)$:
$$\mathbf{x}(t) = \mathbf{H}\mathbf{s}(t) \implies \hat{\mathbf{s}}(t) = \mathbf{W}\mathbf{x}(t)$$
Equation 3: Independent Component Analysis (ICA) unmixing
ICA maximizes the statistical non-Gaussianity (negentropy / kurtosis) of unmixed sources $\hat{\mathbf{s}}(t)$, perfectly isolating ocular blinks, cardiac ECG spikes, and muscle EMG tremor into distinct components that can be zeroed before back-projecting clean brain signals.
04. Grand Challenge 06: Zero-Sedation Pediatric Motion Strategies
👶 Motion Artifact Management in Vulnerable Neonates
Unlike adult subjects, neonates and toddlers cannot follow instructions to remain still, and pharmacological sedation is contraindicated due to neurotoxicity risks. Motion mitigation must be built into hardware and signal processing:
🩹 Non-Adhesive Hydrogel Montages
Rigid electrode cups shear off fragile infant skin during restlessness. Soft, conductive hydrogel arrays conform to curvature with zero shear resistance, preserving low contact impedance without abrasive prep.
🌊 Wavelet Spline Reconstructions
In bedside fNIRS, head twitches create sudden sharp steps in optical density ($\Delta\text{OD}$). Discrete wavelet transforms identify coefficients whose absolute values exceed $3.5\sigma$, applying cubic spline interpolation to eliminate baseline jumps without distorting slow hemodynamic waves.
⚡ Doppler Wall Clutter Filtering
In transfontanellar ultrafast ultrasound, large pulsatile vessel wall motion ($v_{\text{wall}} \sim 1 - 5\,\text{mm/s}$) obscures slow microcapillary blood flow ($v_{\text{blood}} \sim 0.1 - 1\,\text{mm/s}$). Singular Value Decomposition (SVD) spatiotemporal clutter filters selectively eliminate high-energy, low-rank tissue motion.