Transcranial Pulse Stimulation with Ultrasound in Alzheimer's Disease—A New Navigated Focal Brain Therapy

Abstract Ultrasound‐based brain stimulation techniques may become a powerful new technique to modulate the human brain in a focal and targeted manner. However, for clinical brain stimulation no certified systems exist and the current techniques have to be further developed. Here, a clinical sonication technique is introduced, based on single ultrashort ultrasound pulses (transcranial pulse stimulation, TPS) which markedly differs from existing focused ultrasound techniques. In addition, a first clinical study using ultrasound brain stimulation and first observations of long term effects are presented. Comprehensive feasibility, safety, and efficacy data are provided. They consist of simulation data, laboratory measurements with rat and human skulls and brains, in vivo modulations of somatosensory evoked potentials (SEP) in healthy subjects (sham controlled) and clinical pilot data in 35 patients with Alzheimer's disease acquired in a multicenter setting (including neuropsychological scores and functional magnetic resonance imaging (fMRI)). Preclinical results show large safety margins and dose dependent neuromodulation. Patient investigations reveal high treatment tolerability and no major side effects. Neuropsychological scores improve significantly after TPS treatment and improvement lasts up to three months and correlates with an upregulation of the memory network (fMRI data). The results encourage broad neuroscientific application and translation of the method to clinical therapy and randomized sham‐controlled clinical studies.

CERAD CTS score analysis (N = 35) revealed a significant within-subjects effect of TIME: P < .0001. The between-subjects effect of CENTER was not significant (P = .313). This indicates that CTS values differ between the 4 time points, but not overall between centers. Post-hoc pairwise comparisons of CTS values (Bonferroni corrected) unveil significant differences for the following: baseline < post-stim (P Bonf < .0001), baseline < 1month post-stim (P Bonf < .0001), baseline < 3months post-stim (P Bonf < .0001; see Supplementary Table S4 and Figure 3A). Furthermore, a significant interaction TIME*CENTER (P = .003) was found indicating that CTS differences between time points vary between the centers. A follow-up repeated measurements ANOVA for both centers separately revealed a significant main effect of TIME for both centers individually. All 3 pairwise comparisons remained significant for center 2, whereas for center 1 only the baseline < post-stim contrast reached significance. 2

CERAD Logistic Regression (LR) score
For the CERAD LR score (N = 31) a significant within-subjects effect of TIME (P < .0001, Greenhouse-Geisser corrected since the assumption of sphericity was violated) was found. In contrast, the between-subjects effect of CENTER was not significant (P = .830) indicating that LR values differ among the 4 time points but not over all between the centers. Post-hoc pairwise comparisons revealed significant differences for baseline < post-stim (P Bonf < .0001), baseline < 1month post-stim (P Bonf < .0001), baseline < 3months post-stim (P Bonf < .0001), post-stim < 1month post-stim (P Bonf = .012; Table S4, Figure 3B). As for the CTS score, a significant interaction TIME*CENTER (P = .038) was found. Further, the main effect of TIME was significant for both centers in a repeated measurements ANOVA for both centers separately. Again, for center 2 all 3 baseline comparisons remained significant, but for center 1 only the baseline < 1month poststim comparison reached significance.

CERAD Principle Component Analysis (PCA)
Three factors achieved eigenvalues greater than 1 which means that they explained more variance than every single subtest taken alone (Table S5). Factor 1 (eigenvalue = 5.09, explained variance = 46.25%) displayed the highest loadings on the delayed recall and recognition of the Word List and on Savings of the Word List and the Figures and was thus named Factor MEMORY. Factor 2 (eigenvalue = 1.53, explained variance = 13.95%) was interpreted as VERBAL as its highest loadings were found for the Verbal Fluency tasks and the Word List Total. The loadings of Factor 3 (eigenvalue = 1.19, explained variance = 10.77%) were highest for the figural tasks and this factor was termed FIGURAL.

Factor 2 (VERBAL)
Again, the mixed ANOVA showed a significant within-subjects effect of TIME: P < .0001 (Table   S4, Figure 3D). The between-subjects effect of CENTER was not significant (p = .137). Post-hoc pairwise comparisons demonstrated significant differences for baseline < post-stim (P Bonf = .003), baseline < 1month post-stim (P Bonf = .001), baseline < 3months post-stim (P Bonf < .0001). Further, a significant interaction TIME*CENTER (P = .002) was found indicating that factor differences between time points vary between centers. The main effect of TIME was significant for center 2 only (repeated measurements ANOVA for the centers separately) and all 3 pairwise comparisons remained significant.

Factor 3 (FIGURAL)
Factor 3 revealed a significant within-subjects effect of time in the sense of a decline (P = .014; Table S4, Figure 3E). The between-subjects effect of CENTER was not significant (P = .165). Posthoc pairwise comparisons showed a significant decline for baseline > 3month post-stim (P Bonf = .007). In addition, a significant interaction TIME*CENTER (P = .015) was found demonstrating that factor differences between time points differ between centers. The main effect of TIME was significant for center 1 only and the decline for baseline > 3month post-stim remained significant (repeated measurements ANOVA for the centers separately). A further qualitative evaluation indicates that this effect is primarily due to a considerable decline of constructional praxis (Figures -Copy) of the patients of center 1. 4

Depression scores cannot explain neuropsychological improvement
For the GDS, the effect of TIME was significant (P = .005). Pairwise comparisons (Wilcoxon-Test) showed GDS improvement for baseline > 3months post-stim (P Bonf = .012). For BDI, effect of TIME was also significant (P < .0001). Pairwise comparisons (Wilcoxon-Test) displayed BDI improvement for baseline > post-stim (P Bonf = .012), baseline > 1month post-stim (P Bonf = .006) and baseline > 3months post-stim (P Bonf = .012). Importantly, there was no significant correlation between BDI / GDS scores and global CERAD scores (CTS, LR) or the PCA factors after accounting for multiple comparisons (Bonferroni correction). This indicates that CERAD improvements were not driven by changes of depressive symptoms.

Improvement of subjective patient performance
Results from post treatment standard scales showed significant improvements in the subjective evaluation of memory performance (SEG) over time (within-subjects effect of time: P = .027, pairwise comparisons not significant). The other standard scales did not show significant changes.
In the post-treatment questionnaires, up to 20% of the patients reported subjective improvements and only 2-3% aggravations (details in Table S6).

TPS data simulations
Temporal peak intensity fields as generated by the clinically applied TPS system have been simulated for free degassed water and two real skulls including brain tissue. The numerical models were reconstructed from the CT scans of the complete heads of two donors. The position of the TPS source in relation to the skull was recreated, according to the configuration of the experimental measurements described below. The numerical simulations were performed using Matlab 5 (Mathworks, USA) and the open-source k-Wave toolbox, which uses a k-space pseudo-spectral time domain solution to coupled first-order acoustic equations. [1] The simulation was limited to a volume (98x50x50 mm³) of the head containing the expected focal area and the surroundings ( Figure S2A) as extracted from the CT scans. The Hounsfield Units were converted into density and acoustic celerity using the built-in k-wave functions based on the empirical results of Schneider et al. and Mast. [2,3] Absorption coefficients of 3.57 dB.cm -1 .MHz -1 and 0.58 dB.cm -1 .MHz -1 were respectively assigned to bone and brain structures (Szabo 2014). [4] The region outside of the skulls was modeled as non-absorbing water (ρ = 1000 kg.m -3 , c = 1489 m.s -1 ). The non-linearity parameter B/A was set to 7, corresponding to most of biological soft tissues including brain, [4,5] for the whole computational domain. The pressure source was modeled as a brass parabolic reflector (c = 4198 m.s -1 ; ρ = 8470 kg.m -3 ) centered on a cylindrical coil, matching dimensions those of the real device.
The initial acoustic excitation was simulated as a cylindrical pressure wave uniformly distributed over the coil and modelled as a single-pulse.

Human skull and brain sample measurements
Single pulse pressure waves were generated by a device with the same acoustic performance as the system used for the clinical study (Storz Medical AG, Tägerwilen, Switzerland). A typical pressure pulse generated by this device, measured at the focus, is shown in Figure 1B and the experimental setup is illustrated in Figure S1. The pressure pulses were measured using a needle hydrophone (Dr. Müller Instruments, Oberursel, Germany) fixed on a two-axis sliding stage. The predefined measurement domain (50 mm along the beam axis and 40 mm along the transversal axis) was centered on the geometrical focus of the handpiece, defined as the origin of the coordinate system.
The spatial transversal and axial measurement steps were kept below 1 mm and 3 mm respectively.
All pressure waves were released at a drive level of 0.25 mJ/mm 2 , and at a pulse repetition frequency of 2 Hz. First, a reference acoustic field measurement was performed in free water. Then, a section of human skull (roughly intermediate between bregma and lambda), with the brain 6 parenchyma completely removed, was placed in front of the handpiece and firmly fastened in a holder. The relative position of the handpiece to the skulls was determined from photographic acquisitions during the measurements. A 3D reconstruction of the TPS handpiece and the mounting plate of the water basin was recreated in Blender software [https://www.blender.org/]. The CT scans were segmented to create a 3D surface model of the human samples. Virtual cameras using the specifications of the Canon EOS 5D Mk II were then aligned and positioned to match the reference images. The plane of incision around the circumference of the skull was determined to create a 3D model of the skull section, which was then used to reconstruct the measurement setup. These steps allowed a discrete transformation between the CT image system and real-world geometric focus position. The CT data was transformed and interpolated to the resolution of 200um, which allowed for an easy extraction of both the 3D computational volume for the simulations described above and the image plane for the visualization of the 2D measurements. The measured fields were displayed in the corresponding slice in such a way that the origin of the coordinate system of the measurement setup, representing the geometrical focus, matches the corresponding position in the CT slice ( Figure S2). A similar procedure was used for measuring the pressure drop in 10 human brain samples in vitro (brain soft tissue stabilized with a net construction but without skull, 0-7 days post mortem).

Rat skull measurements
For allowing judgements about differences between animal skulls and human skulls, we also performed measurements of a rat skull with the same principal technology. For this, TPS was applied with 0.1 mJ/mm 2 , 0.35 mJ/mm 2 and 0.55 mJ/mm 2 and at 6 positions of the rat skull: bregma point, about 5 mm left and right from the bregma, lambda point and about 5 mm left and right from the lambda ( Figure S2C). Attenuation of the pulse intensity was measured at peak pulse intensity below the skull with a needle sensor (Müller-Platte needle hydrophone, Dr. Müller Instruments, Oberursel, Germany). Pulse amplitude was measured at the TPS focus below the skull.

TPS safety investigations in anesthetized rats
Single pulse pressure waves were generated by a device equivalent to the TPS system in the clinical study in 2 rat studies (Storz Medical AG, Tägerwilen, Switzerland). TPS was applied at a fixed position over the rat skull and a constant focus in the brain under anesthesia. Study 1 used 80 male In more detail, pulse energies in study 1 were varied and sonication was performed at a frequency of 3 Hz in groups of 10 rats each -1 control and 7 test groups with the following settings: 0.1 mJ/mm 2 , 100 pulses; 0.1 mJ/mm 2 , 200 pulses; 0.1 mJ/mm 2 , 400 pulses; 0.2 mJ/mm 2 , 100 pulses; 0.2 mJ/mm 2 , 200 pulses; 0.2 mJ/mm 2 , 400 pulses; 0.3 mJ/mm 2 , 100 pulses. For safety evaluations, brain preparations (80 rats) and histological investigations (16 rats) were performed to investigate for possible intracerebral bleeding and tissue damage as primary outcomes. Outside the safety context of this study, animal behavior was also analyzed. Rats were held in groups in Makrolon-IV cages at fixed climatization and 12h:12h light-darkness cycles. For anesthesia isoflurane 1-2% and fentanyl 5µg/kg or Butorphanol 3,3 mg/kg were used. Analgesia was required for controlled experimental conditions and a stable brain stimulation focus. For postsurgical analgesia carprofen was used. For post treatment brain preparations, animals were decapitated. Study 2 used a similar setting and evaluated the sonication effects via in vivo MRI (see Figure S3). Total energy dose in 5 rats was varied between 400, 4000 and 8000 pulses (at 0.2 mJ/mm 2 ) corresponding to 15-, 150-and 300-fold energy levels relative to the human dose allowed with the certified TPS system. Figure S1. Experimental setup for the skull and brain sample measurements. The TPS handpiece was fixed on the side of a basin filled with degassed water. Test specimens (e.g. human skulls, rat skulls, brain specimens) were fixed directly in front of the handpiece. For brain specimen fixation a net was used.

Supplementary Figures
Specimen related pulse attenuations were recorded by the Hydrophone with reference to free water results (compare figure S2). The Hydrophone can be moved in 3D.   Tables   Table S1. Correlation between       18