Auditory Mondrian masks the airborne-auditory artifact of focused ultrasound stimulation in humans

Low-intensity Focused Ultrasound Stimulation (FUS) can modulate neural activity in cortical, subcortical, and deep brain regions, achieving millimeter precision through transcranial ultrasound stimulation (TUS) and can affect behavior [1, 2]. However, there is concern that FUS may induce an auditory effect, cortically activating the subject’s auditory sensation, thus confounding behavioral and electrophysiological responses during TUS research. [3–7]. Several studies have mitigated this auditory artifact through ramping [8] or audio masking [5, 6], but prior research has had two main limitations. First, previous studies addressed a limited range of the total FUS parameter space: fundamental frequency (f0), pulse repetition frequency (PRF), duty cycle (DC), sonication duration (SD), and intensity (I) [1, 7, 9–11]. Second, as these studies used direct stimulation of subjects, they did not distinguish between airborne or tissue conduction effects [5–7]. Here, we evaluate the airborne auditory artifact over a range of FUS parameters through sonographic characterization, the human subject’s response to recorded audio clips, and a two-interval forced choice (2IFC) task to test the effectiveness of three mask types: square [5, 6], pulsed sine, and random multitone. The multitone random mask, or Auditory Mondrian, is inspired by the visual Mondrian used in the continuous flash suppression to mask visual targets [12].

We recruited 228 healthy participants for the three online auditory psychophysical experiments (See Supplementary Methods for details). In experiment 1, participants performed a detection task in which they were asked whether they detected a distinct sound while listening to audio recordings of FUS sham and stimulation trials. In experiments 2 and 3, participants performed a two-interval forced choice (2IFC) task in which they chose which interval of a pair contained the FUS stimulation embedded in an auditory mask. Audio clips from the microphone were used without volume (loudness) manipulation and confirmed by experimenters to match the sound produced from the FUS setup.

In an artificial environment, we found that the ultrasound transducer is a primary source of airborne auditory artifacts (Fig. S1 A and B). Short-time Fourier transforms (STFT) of the audio recordings of FUS revealed clear frequency bands at the PRF and harmonics, along with additional frequency bands in the human hearing range that did not fit with the corresponding PRF (Fig. 1 A). These additional frequency bands were consistent at approximately 8 and 12 kHz throughout all PRF and even seen with continuous wave US bursts (Fig. 1 A, yellow arrows) and appeared regardless of the coupling method or the cone and arm setup (Fig. 1 B). The electrical spectrum density showed no peaks at the human frequency range, so it likely did not contribute to the auditory artifact. Based on the above acoustic analyses, we concluded that the ultrasound transducer is a source of airborne auditory artifacts.

Figure 1. Auditory Mondrian effectively masked the auditory artifact from both continuous and pulsed FUS.

(A and B) Acoustic analysis of auditory artifact with FUS. (A) Short-time Fourier transforms (STFT) of the audio clips were recorded when we stimulated with pulsed ultrasound bursts (F0 = 270 kHz, SD = 500 ms, PRF indicated in kHz, DC = 50%). STFT showed clear frequency bands at the corresponding PRF and their harmonics, along with additional frequency bands (approximately 8 and 12 kHz, yellow arrows) in the human hearing range. (Right) STFT of a continuous ultrasound burst (F0 = 270 kHz, SD = 500 ms) when the FUS setup was coupled with ultrasound gel to the water tank (Coupled) or directly immersed in the tank (Direct). (B) STFT of a continuous ultrasound burst (F0 = 270 kHz, SD = 500 ms) when the FUS setup was coupled to the water tank using two different cones and metallic arms (Setup 1 and 2). The frequency bands in the human hearing range appear at approximately 8 and 12 kHz (yellow arrows). (C) The design of the Two-interval forced choice (2IFC) tasks. (D) Properties of the FUS burst used to produce the auditory artifact and the audio mask properties. (E and F) Results from the 2IFC (experiment 3) comparing the square and the Auditory Mondrian masks. Box plot of the percentage correct response (upper panels) and the percentage confidence rating (lower panels) for each mask overlapping with increasing pressures (0.4 – 1.2 MPa with (E) the continuous FUS waveform and (F) the pulsed waveform. Red lines indicate the 50% chance level. Red asterisks indicate paired t-test significance from chance level with Bonferroni corrected p < 0.001. Error bars indicate standard deviation.

In the detection task (experiment 1), participants significantly detected the auditory artifact (see supplementary audio tones 1 and 2 and their STFT in Fig. S2–3) with high confidence from the sham at nearly all tested parameters (Fig. S4, paired t-test from chance, Bonferroni corrected p < 0.001). For continuous US bursts (Fig. S4 A–C), an F0 of 375 kHz was significantly undetected compared to other F0, likely due to diminished transducer efficiency (one-way repeated measures ANOVA F_{(3, 147)} = 100.672, p = .000, with Tukey post-hoc test p < 0.5). An SD of 125 ms was the most perceivable, with a gradual decline as the duration increased (one-way repeated measures ANOVA F_{(4, 196)} = 13.966, p = .000, with Tukey post-hoc test p < 0.5 between 50 ms and 125 ms). Detection increased linearly with increasing pressure (one-way repeated measures ANOVA F_{(2, 98)} = 119.327, p = .000, with Tukey post-hoc test p < 0.5 between P = 0.4 MPa and other pressures). For pulsed US bursts (Fig S4 D–I), there was a significant decline at PRF = 20 kHz, the upper range of the human hearing range, in both detection level and confidence (d prime, one-way repeated measures ANOVA F_{(7, 343)} = 35.634, p = .000, with Tukey post-hoc test p < 0.5 between PRF = 20 kHz and other frequencies).

In the 2IFC tasks (experiments 2 and 3), human participants testing confirmed that Mondrian outperformed Monotone audio masks at masking airborne auditory artifacts (Fig. 1 C and D). We tested the effectiveness of a square mask (1 kHz, similar to previously reported [5, 6]), a pulsed sine mask (PF= 7 kHz, PRF = 1 kHz, and DC = 50%, simulating the FUS), and a multitone random mask (Auditory Mondrian) consisting of eight overlapping mini-pulsed-sine tones (SD = 500 ms, PF = 15 kHz, PRF = 1 – 15 kHz, DC = 50%) with random shifts and a total duration of 2 sec (see supplementary audio tones 3–12 and Fig. S5). The square and pulsed sine masks decreased participant confidence to chance level at all continuous and only low-pressure pulsed waveforms (Fig. S6, paired t-tests from chance with Bonferroni corrected p > 0.05). In contrast, the Auditory Mondrian mask decreased participants’ confidence to chance level at pulsed FUS at all tested pressures (Fig. 1 E and F, paired t-tests from chance with Bonferroni corrected p > 0.05).

In conclusion, we identified a source of airborne auditory artifacts in FUS arising from the ultrasound transducer, with some contribution from the holding apparatus, generating tones within the auditory spectrum that are both dependent and independent of PRF. Human detection of the auditory artifact depended on the US parameters of I and F0 for the continuous waveform and PRF for the pulsed waveform. Finally, an Auditory Mondrian effectively masked these airborne auditory artifacts. This study has several limitations. First, we only focused on the auditory artifact’s airborne component rather than the tissue-conduction component [6] of the auditory artifact.

Nevertheless, this work will aid further researchers in isolating and excluding the airborne part. A second limitation is our use of online psychophysics experiments. While in-person would be ideal, modern online auditory psychophysics experiments have been proven to match in-person results [13, 14]. This study introduces the Auditory Mondrian as a new masking design for acoustic psychophysics [15]. As a new design, further investigation and parameter adjustment with in-person direct FUS stimulation studies are needed to maximize its efficiency while minimizing unpleasant auditory perceptions.