Understanding intracranial aneurysm sounds via high-fidelity fluid-structure-interaction modelling

Intracranial aneurysms are common and rupture carries high morbidity and mortality despite generally low annual rupture risk. There is a clinical need for better rupture risk indicators beyond systemic and morphological factors. Prior hemodynamic studies have not yielded universally accepted predictors, but flow instabilities have been identified in some aneurysms and hypothesized to induce wall vibrations that may drive deleterious remodeling or rupture. Aneurysm sounds (bruits and musical murmurs, ~100–800 Hz) have been reported for decades, yet their mechanisms remain debated: self-excitation by pulsation, transmission of vortex shedding or turbulent flow, or structural resonance of the aneurysm wall. Mathematical models have often assumed simplified geometries and breathing-mode vibration, typically neglecting realistic internal flow. Recent computational and in vitro work suggests whole-sac rocking modes may occur. This study aims to elucidate the mechanisms, magnitude, and frequency content of aneurysm wall vibrations under physiologically pulsatile flow using high-fidelity fluid-structure interaction (FSI), and to relate these vibrations to clinically observed bruits and murmurs.

Earlier clinical and experimental observations documented intracranial aneurysm sounds in the 100–800 Hz range and proposed mechanisms including self-excitation due to pulsation, transmission of turbulent or vortex-shedding flow, and structural resonance. Analytical models of aneurysm resonance often presumed spherical sacs and breathing modes, simplifying geometry and flow. Prior FSI studies primarily captured low-frequency wall motion from pressure pulsation using numerics that likely damped flow instabilities, thereby not reporting high-frequency vibrations. More recent CFD has identified flow instabilities in bifurcation aneurysms, and proof-of-concept high-fidelity FSI under steady or ramped inflow showed high-frequency vibrations and rocking modes but did not simulate full physiological pulsatility. There remains a gap in mechanistic understanding of how pulsatile flow instabilities excite structural modes and generate the spectral signatures of bruits and musical murmurs observed clinically.

Study design and cases: Six middle cerebral artery aneurysm cases from a previously published cohort were studied: five bifurcation aneurysms previously exhibiting flow instability (Cases 3, 9, 11, 12, 16) and one sidewall aneurysm as a stable-flow control (Case 8). Ages spanned 30s–80s; four of six were female. Ethics approvals and consent for the original anonymized imaging cohort were in place; secondary anonymous data use was waived. Boundary conditions: Physiologically pulsatile inflow was prescribed by scaling an internal carotid artery (ICA) waveform (older adults) by cycle-averaged flow, damped 10% to approximate transit to MCA. Mean ICA cross-sectional average velocity was 0.37 m/s; flow rate scaled with inlet area. Internal arterial pressure varied between 80–120 mmHg; with 10 mmHg intracranial pressure subtracted, the applied internal pressure ranged 70–110 mmHg, following the same waveform shape as inflow. A smooth sinusoidal initialization (first 0.25 s) matched value and slope to avoid transients. Inlet velocity profiles were fully developed Womersley; outlets had zero-pressure (do-nothing) conditions. Structural domain and deformability: The deformable region encompassed the sac and nearby branches, defined by a sphere centered at the sac centroid with at least 0.5 mm clearance; proximal and distal inlet/outlet segments outside this sphere were rigid and fixed. Material properties: A hyperelastic St. Venant–Kirchhoff wall model with E = 1×10^6 Pa and ν = 0.45 was used. Blood was Newtonian with μ = 0.0035 Pa·s and ρ = 1000 kg/m^3. Wall thickness was set to 0.25 mm. Geometry preprocessing and meshing: Because medical image-based geometries are acquired in a pressurized state, a zero-pressure geometry was inferred by pressurizing the as-segmented model to the cycle-averaged pressure (82 mmHg) and inverting deformations so that the pressurized computational geometry matched the original segmentation. Fluid domains had ~100,000 tetrahedral elements (average edge length 0.234–0.371 mm) with two boundary layers totaling 0.25 mm. At least 10 inlet diameters of length were included with 3-diameter cylindrical extensions to ensure circular inlets/outlets. Solid domains had 35,000–45,000 elements with two additional layers through the 0.25 mm wall thickness. FSI solver and numerics: Simulations used turtleFSI (monolithic, fully coupled), semi-implicit shifted Crank–Nicolson time integration (θ=0.501), and Newton iterations (absolute tolerance 1e−10, residual tolerance 1e−9). Elements were P2–P2–P1 (third-order spatial accuracy). Timestep was 0.340 ms (≈2800 steps per 0.951 s cardiac cycle). No turbulence model was used. Each run used 40 Intel Skylake cores on Niagara HPC; compute times were ~32–95 h per case. Six cycles (5.6 s) were simulated; analyses focus on the fourth cycle after demonstrating cycle-to-cycle convergence. Signal processing and metrics: A 25 Hz threshold separated low-frequency inflation/stable flow (<25 Hz) from high-frequency vibration/flow instability (>25 Hz), capturing >99% of cardiac waveform power below 25 Hz. High- and low-pass filtered Q-criterion fields illustrated unstable versus stable vortical structures. To characterize spatial distribution and scale, windowed RMS amplitudes (rectangular window, 250 timesteps) were computed for each vector component of high-pass filtered wall displacement ("vibration amplitude") and fluid velocity ("flow instability amplitude"). The combined magnitude and the spatial 99th percentile were used as robust summary measures over time. Spectral analysis: Short-time Fourier transforms were computed over a 1.051 s window (one cycle plus preceding 0.1 s), using 7 windows with 75% overlap (1240 timesteps per window), yielding frequency resolution 2.87 Hz and time resolution 0.105 s. Component-wise spectrograms (X, Y, Z) were averaged to obtain representative fluid velocity and wall displacement spectrograms. Mode extraction: For displacement bands extending beyond fluid spectral content at the same frequencies, band-pass filters (20–40 Hz bandwidth) isolated structural modes. The remaining broadband "bruit" was obtained by band-stop filtering to remove identified modal bands. Vibration mode shapes and amplitudes were visualized and compared across cases.

• Flow instability and vibrations: In five cases (3, 9, 11, 12, 16) with unstable flow, wall vibrations were present; the stable-flow control (Case 8) showed only very low self-excitation at 25–35 Hz. Unstable cases showed broad-band, random high-frequency vibrations (bruits) and distinct narrowband resonant bands (musical murmurs) consistent with structural modes.
• Amplitudes: Both bruits and modes had similar orders of magnitude across cases, typically 0.1–1.0 μm. Ranking by overall vibration amplitude: Case 3 highest (~1.3 μm), then Cases 9 (~0.6 μm) and 16 (~0.5 μm), with Cases 12, 11, and 8 lowest. Peak vibration RMS amplitude reached ~4% of total wall displacement in Case 3; pressure-driven inflation dominated overall displacement.
• Spectral characteristics: In fluid spectra, transitional/turbulent broad-band content appeared during systole and dissipated in diastole; only Cases 3 and 9 showed weakly repeating harmonic bands suggestive of vortex shedding. Wall displacement spectrograms in unstable cases displayed one or more narrow bands in the hundreds of Hz that persisted into diastole after flow instability decayed; these did not exhibit strong harmonic repetition typical of vortex shedding.
• Modal frequencies and shapes: Extracted bands corresponded to whole-sac structural rocking modes. Mode 1 commonly involved rocking between daughter branches; higher modes showed rocking along other axes or whole-sac "bouncing." Modal frequencies inversely related to aneurysm size: larger sacs (e.g., Case 16) had lower bands (~120–190 Hz), smaller sacs (e.g., Case 9) higher bands (~360–495 Hz).
• Bruit vs mode amplitudes: Relative contributions varied by case. Examples: Case 3 mode amplitude ~1.3 μm versus bruit ~0.5 μm; Case 16 bruit ~0.4 μm versus first mode ~0.3 μm; in Cases 9, 11, 12 the bruit amplitude exceeded individual modes.
• Correlations: 99th percentile vibration amplitude correlated strongly with 99th percentile flow instability amplitude (r^2 ≈ 0.81). Stable-flow velocity magnitude (<25 Hz) was less predictive (r^2 ≈ 0.63). Aspect ratio showed a moderate association (r^2 ≈ 0.59), while sac volume was not predictive (r^2 ≈ 0.18).
• Clinical concordance: Power spectra and spectrograms resembled prior patient recordings. Cases exhibited patterns analogous to normal (no peaks), single-peak, or multi-peak spectra (100–500 Hz), similar to Kurokawa et al. Murmur bands persisting into diastole paralleled Doppler observations by Aaslid and Nornes, supporting the interpretation of structural resonance rather than pure vortex-shedding harmonics.

The simulations indicate that aneurysm wall vibrations predominantly arise from flow instabilities rather than pressure pulsation alone and comprise two components: broadband transmission of turbulent/transitional bruits and narrowband structural resonance (rocking modes). The narrow bands that extend into diastole match the clinical description of musical murmurs and, in contrast to classic vortex-shedding signatures, do not show strong harmonic repetition; their diastolic persistence reflects structural resonance that outlasts the driving flow instability. Frequency content and temporal evolution over the cardiac cycle align with previously reported microphone and Doppler ultrasound recordings, lending credibility to the mechanistic interpretation. Differences in peak sharpness and frequency compared to in vivo recordings likely reflect unmodeled damping and constraints from surrounding tissues (brain, CSF, bone), which would increase effective stiffness, broaden peaks, and shorten ring-down. Mechanistically, higher aspect ratio and narrow necks (e.g., Case 3) may confer greater susceptibility to rocking due to lower bending resistance at the neck. The strong correlation between flow instability amplitude and vibration amplitude supports a causal link, suggesting that assessing flow instability may help infer vibration propensity. Given experimental and clinical evidence that high-frequency mechanical stimuli can induce endothelial dysfunction or alter smooth muscle behavior, the observed vibration environment could be a relevant mechanobiological factor in aneurysm remodeling and possibly rupture risk. The study provides quantitative vibration amplitudes and frequencies to guide in vitro investigations into cell-level responses, and motivates further exploration of vibrations as a potential biomarker or therapeutic consideration.

This work presents the first high-fidelity, pulsatile FSI simulations demonstrating that intracranial aneurysm sounds arise from a combination of broadband bruits due to flow instability and narrowband structural resonances (rocking modes) of the aneurysm sac. Vibrations in unstable-flow aneurysms had amplitudes on the order of 0.1–1 μm, with modal frequencies primarily in the 100–500 Hz range and inversely related to sac size; stable-flow aneurysms showed negligible vibration. The spectral and temporal features of the simulated vibrations agree with previously recorded intracranial bruits and musical murmurs, providing a plausible mechanistic explanation and quantitative characterization of the aneurysm wall’s vibratory environment. Future research should quantify how such vibrations affect vascular wall biology, incorporate more realistic tissue constraints and damping, and pursue preclinical validations and imaging-informed models to better relate vibrations to wall pathology and rupture risk.

Models assumed uniform wall thickness, near-linear hyperelastic material behavior, and no viscoelasticity, as well as the absence of surrounding tissues or fluid (brain, bone, CSF) and potential peri-aneurysmal contact; such factors would alter effective stiffness, damping, and modal frequencies/amplitudes. Wall thickness and material heterogeneity, known in real aneurysms, were not represented and could localize or change vibration transmission. Generalized but physiologically plausible flow rate and pressure waveforms were used rather than patient-specific boundary conditions; the study aimed to reveal phenomenology rather than patient-specific predictions. Preliminary tests with more nonlinear material models and CSF damping modified modal frequencies and ring-down duration but did not change the overall phenomena. The sample size was small (n=6) and limited to MCA aneurysms, which may affect generalizability.