In commercial systems for DBS treatment, stimulation parameters range from 0 to 25 mA of amplitude, from 10 to 450 μs of pulse-width, and 2 to 500 Hz of frequency, provided both in monopolar and bipolar fashion (Paff et al., 2020). Empirical observations showed that, in PD, clinical benefits can be fully achieved with a narrower parameters space. For instance, when considering patients with PD, tremor, bradykinesia, and rigidity progressively improved between 2 and 3 V and did not continue to improve beyond 3 V (Moro et al., 2002) that, for an average monopolar impedance of 1,000 Ω, is equal to 3 mA. In the absence of lead damages, for platinum-iridium electrode with area of 0.06 cm², impedances may vary between 500 and 2 KΩ (Kuncel and Grill, 2004). In clinical practice, the amplitude threshold for inducing a clinical response or side effect for each electrode contact is determined by using monopolar stimulation and a stepwise increase in amplitude of 0.2–0.5 V (0.2–0.5 mA) (Volkmann et al., 2002), thus requiring a minimum amplitude resolution of 0.2 mA. For STN stimulation, a 60-μs pulse width is generally used because of its neurons’ chronaxie, and it was empirically observed as being effective on rigidity and bradykinesia (Moro et al., 2002). Lowering the pulse width helps in augmenting the therapeutic window based on the intensity-pulse duration chronaxie relationship (Reich et al., 2015). Frequency stimulation above 200 Hz (Moro et al., 2002) did not show any notable improvements, whereas frequencies below 50 Hz generally worsen Parkinsonian symptoms (Wojtecki et al., 2006).

Despite specific parameters choice, because the electrical safety of the stimulation has to be guaranteed, intrinsic constrains depend on the material and geometries of the electrodes. Stimulation waveforms shall be charge balanced, in active/or passive manner for preventing electrode and tissue interface damage (Cogan et al., 2004). Moreover, intensity and pulse width combination shall be controlled, on the basis of electrode surface, to avoid excessive charge density injection per phase (Merrill et al., 2005). In particular, for conventional platinum-iridium electrodes (i.e., Model 3389, Medtronic), the limit for charge density is 30 μC/cm²/phase.

In AlphaDBS System, the narrower parameter space (frequency of 50–200 Hz, pulse width of 40–250 μV, and amplitude of 0–5 mA), with charge balanced waveforms and with specific controls allowing to reliably guarantee electrical safety, is considered as stimulation requirement for AlphaDBS, without introducing any specific innovation in the stimulation module, which is an established technology. The parameters space, however, could be suitable also for other potential DBS applications (e.g., dystonia or tremor). The sensing and data management modules are the places where innovation for bidirectional neural interfaces is needed.

To implement aDBS, clinical IPGs need to record artifact-free neural activity during stimulation delivery.

Although external systems were able to solve the artifact rejection problem (Rossi et al., 2007; Arlotti et al., 2016b; Petkos et al., 2019), the size and power constrains of implantable operations make the rejection of stimulation artifact a technical implementation challenge. The stimulation artifact consists of direct components (stimulus time-locked voltage transients) and indirect components (voltage decay in the inter-pulse period) (Zhou et al., 2019). Direct artifacts at the adjacent recording electrodes are in the order of volts (common mode artifact) or hundreds of millivolts (differential mode artifact). In DBS applications, the differential artifact amplitude imposes a minimum input range of 100 mVpp, but real-world impedances mismatch may lead to greater values. The sensing module should avoid saturation for differential artifact greater than 100 mVpp while resolving 1-μV signals. In fact, as reported in the literature, implantable DBS devices with sensing capabilities (i.e., Medtronic Activa PC + S) are not able to provide artifact-free meaningful recordings (Cummins et al., 2021). Symmetric electrode configuration (Figure 1A) with input blanking has been applied as means to mitigate the differential and common mode artifacts (Stanslaski et al., 2018), providing better artifact management (Cummins et al., 2021) but introducing a limitation in choosing the best stimulation configuration for the patient.

Any combinations of electrode contacts should be selectable for recording with respect to the stimulation one (Figure 1). In the absence of this requirement, given a conventional quadripolar linear DBS lead, the selection of the extreme contacts would deny recording possibilities, because of the unavailability of recording contacts symmetrical to the stimulation one. Even worse, for directional leads (Eleopra et al., 2019), the electrode position together with the different area of directional contacts vs. cylindric contacts would lead both to unbalanced impedances and spatially asymmetrical recording contacts, thus deteriorating artifact rejection.

In case of clinical closed-loop DBS, a stimulation agnostic sensing module, able to reject the stimulation artifact independently from the stimulation shape and configuration (monopolar and bipolar), is preferrable to freely set the most effective stimulation. The possibility to be stimulation agnostic depends on the artifact rejection strategy. For instance, employing input blanking techniques, as described in Stanslaski et al. (2018), limits the choice of the stimulus shape. Ideally, the pulse waveform should be actively charge balanced and symmetrical to minimize the time duration of the stimulus artifact and maximize the benefit of blanking. Conversely, in monophasic passively charge-balanced stimulation, the voltage decay of a single pulse may last for hundreds of microseconds, thus requiring to increase the duration of blanking and data loss.

An optimal sensing module should be also software needless, not requiring for additional software for artifact mitigation or removal. Back-end software solutions have been proposed and implemented in other devices ranging from interpolation (Zhou et al., 2019), support vector machines (Stanslaski et al., 2018), and template matching (Qian et al., 2017). Real-time processing on implantable devices requires computational power, which, as a rule of thumb, should be minimized, but as long as back-end solutions prove to be compatible with low-power real-time processing constrains, they can still be employed.

Therefore, the requirements of the AlphaDBS sensing module to implement a bidirectional deep brain neurostimulator are (1) resolving 1 μV LFPs signals, (2) eliminating differential stimulus artifact (>100 mVpp) and common mode stimulus artifact (>1V), (3) being stimulation agnostic, (4) being electrode configuration independent, and (5) being needless for back-end processing. Requirements (3), (4), and (5) are innovative with respect to other available options of implantable DBS devices with sensing capabilities, altogether providing a reliable and robust system for artifact-free recordings.

Having the object of accelerating neurophysiological research, a core requirement for a bidirectional IPG acting as a clinical BCI is to store and transmit neural signals. Although chronic data streaming represents a heuristic goal, its practical implementation still needs to overcome important limitations such as high-power demand, consequent fast battery drain, and maintenance of a permanent external receiver link; all these features ultimately add unnecessary burdens for patients. For instance, continuous data streaming with an implantable rechargeable device (Gilron et al., 2021) require the use of a transmitter that has to be continuously worn by the patient. Many bidirectional neuromodulation platforms are targeting chronic wireless communication (Zhou et al., 2019) at the preclinical or investigational stage.

A less power-consuming solution for chronic neural activity monitoring is to collect data in an embedded memory located inside the IPG. Its implementation requires to compress the neural data in their spectral features or any features being relevant under a clinical and neurophysiological perspective. The correct trade-off between the tracking needs and the size of the embedded memory is application specific. For instance, in PD, the beta power time course is linked to daily motor fluctuations of the patients (Arlotti et al., 2018; Gilron et al., 2021), thus suggesting that extracting the beta power band and continuously storing it could provide an efficient clinical monitoring. However, available devices have limited memories that are overwritten if data are not downloaded and therefore have time-limited monitoring capabilities (Jimenez-Shahed, 2021).

Embedding compressed data (i.e., spectral power) requires to have an a priori knowledge of what signal features are significant for the specific disease, but because of the exploratory application of clinical BCI, time domain data are necessary to the discovery of new biomarkers and physiological mechanisms of action. Moving from the concept of chronic monitoring to exploratory recording, the requirement of data wireless streaming can be relaxed by limiting it to on-demand and time-constrained streaming sessions that allow for controlled experimental investigations without burdening the patient.

The AlphaDBS System will therefore implement two innovative features: (1) the continuous real-time streaming and visualization of data to be used in experimental settings and (2) a long-term continuous recording of embedded data, with an innovative download strategy guaranteeing no data loss.

Adapting stimulation in real time requires to process a physiological variable and to calculate a new set of parameters based on a given relationship (proportional/adaptive mode) or a lookup table (digital mode or state machine). In the AlphaDBS System, the chosen requirement is to implement embedded data processing that ensures lower power consumption, better data privacy, and shorter time delays in stimulation changes, compared with external processing that, however, increases flexibility and research applicability (Pulliam et al., 2020).

Therefore, the AlphaDBS System is a fully closed-loop system, with an embedded algorithm that uses recorded LFPs as biomarker and adapts the stimulation amplitude accordingly, without the need of any external processing.

According to the requirement defined above, the AlphaDBS system (Figure 2) consists of four main components: an IPG (AlphaDBSipg), a patient controller (AlphaDBSPat), a physician controller (NWKStation), and an external device for data recording and streaming from externalized leads (AlphaDBSext). These components together implement a distributed data management platform for data recording, processing, streaming, and storing.

The AlphaDBSipg sensing is implemented in two modes: the “embedded” mode and the “streaming” mode. In the embedded mode, the IPG records and stores neural data during chronic treatment delivery (conventional DBS, cDBS, or aDBS) in an embedded not volatile memory. During stimulation (either cDBS or aDBS), the system extracts the power value of a selected frequency band and stores one value for each side every minute, two full spectra (from 5 to 35 Hz, one per side) every 10 min, and two values of the stimulation amplitude every 10 min. The patient controller downloads the data stored in the embedded memory of the IPG at every recharging cycle and stores them in a second not volatile memory having higher capacity. These data can be downloaded to a smartphone or laptop through a Bluetooth connection using dedicated custom application programming interfaces (APIs). At present, embedded data downloaded through the AlphaDBSpat are transferred to an app that implements a fast healthcare interoperability resource (FHIR)-based standard data management (ready for future interoperability) and allows historical data visualization and power spectral analysis (Figure 2B). In the streaming mode, the IPG, on demand, streams data to the physician programmer (NWKStation) that acts as a receiver and, in turn, transmits data via UART-to-USB connection to a smartphone or a laptop, which can be used for data storing and visualization thanks to custom APIs. Figure 2C shows the present implementation of a Python-based graphic user interface (GUI) that receives, visualizes, and saves real-time data.

The AlphaDBSipg has a total volume of 20.96 cc and weight of 32.70 g, with a medical grade rechargeable battery of 200 mA/h, retaining the 90% of the capacity at 2,000 cycles. The size is in line with other rechargeable DBS IPGs, such as Boston Scientific Vercise (volume of 20.7 cc and weight of 33 g) and Medtronic Activa RC (volume of 22 cc and weight of 40 g). The header is compatible with Medtronic DBS lead extensions model 37086, and it can allocate two extensions for a total of 16 independent contacts. The AlphaDBSipg electronic board has circuitry for driving 16 stimulation channels, each of them can be configured independently. The output current for each channel ranges from 0 to 5 mA. Multisite stimulation is possible by keeping the duration and the frequency of the pulses fixed. The stimulation waveform is firmware selectable, with both active and passive charge balancing available. In case of active return, the ratio between the cathode and the anode current amplitude is 5, leading to a balancing anodic pulse lasting five times the cathodic one. The AlphaDBSipg has been configured for providing capacitive coupled active charge balanced asymmetric pulses, with frequency ranging between 40 and 200 Hz and pulse width ranging between 40 and 250 μs, despite that frequency can be extended to 2.5 KHz and pulse width to 1 ms. At 5 mA and 250 μs, the charge density injected per phase is 20 μV/cm²/phase, when considering a platinum-iridium electrode having a surface of 0.06 cm² (i.e., Model 3389, Medtronic, Inc.) and the maximum charge density injection accepted is 30 μC/cm²/phase.

The AlphaDBSipg can deliver DBS both in the conventional mode (cDBS), in which stimulation parameters are set using the physician controller and remain fixed, and the adaptive mode (aDBS). In this last case, stimulation amplitude, pulse width, and frequency are dynamically changed on the basis of the embedded closed-loop logic or pre-set via physician programmer and radio frequency (RF) communication. The closed-loop logic implementation now tested is based on the linear proportional feedback mode that uses the LFP beta band (10–35 Hz) as neurophysiological biomarker (Arlotti et al., 2018; Guidetti et al., 2021). In summary, the specific personalized beta band of the patient is chosen by inspecting recorded LFPs using the streaming mode. Then, both the personalized beta band and the therapeutic window are set in the physician controller. When aDBS is ON, the recorded beta band is analyzed and the DBS amplitude is modulated linearly between the maximum and minimum amplitudes set as therapeutic window (Arlotti et al., 2018; Prenassi et al., 2021). The closed-loop logic is fully embedded and implemented at the microcontroller firmware level and does need for external units.

The inputs of two differential sensing channels can be multiplexed, respectively, on any of the eight contacts of each lead, and no blanking technique is used for artifact mitigation. The neural signals, analogically filtered for artifact suppression, are digitally converted by the analogue to digital converter (ADC) at a frequency of 512 Hz. Residual harmonic artifacts, when no completely suppressed, require for a sample frequency being greater than the double of the stimulation frequency to avoid aliasing. No additional digital signal processing for artifact removal is needed. The firmware is fully employed for feature extraction and closed-loop logic implementation. This patented sensing technology was (Priori et al., 2005) already implemented and illustrated in external devices (Rossi et al., 2007; Arlotti et al., 2016b) that were used to collect preliminary data on aDBS in more than 40 patients (Rosa et al., 2015, 2017; Arlotti et al., 2018, 2019; Bocci et al., 2021; Prenassi et al., 2021).

A 2.4-GHz ISM/SRD chip allows data streaming of two sensed signals for a distance up to 10 m, firmware upgrade, and bidirectional communication with the patient and the physician controller. The firmware is upgradable through an on-air boot loading functionality.

The AlphaDBSipg has two different microcontrollers: one dedicated to the sensing module, and one to manage the stimulation module, the battery functions, and the RF streaming.

The AlphaDBS System received european (CE) mark for conventional DBS and sensing in January 2021.

Here, we report the results of LFP recordings in the first three patients implanted with the AlphaDBS System and involved in the pilot study NCT04681534. All patients were previously implanted with Medtronic 3389 electrodes, having four cylindrical contacts per side (left side: contacts 0-1-2-3, where contact 0 represents the most ventral and contact 3 represents the most dorsal; right side: contacts 8-9-10-11, where contact 8 represents the most ventral and contact 11 represents the most dorsal).

Data streamed showed limited data loss (average 2%) and no cardiac artifact in any recordings (Table 1). The signal-to-noise ratio (SNR) calculated on the beta peak was always greater than three logs, suggesting an optimal recording performance (Table 1). Representative raw LFPs and the correspondent PSDs obtained during data streaming are reported in Figure 3. As shown in Figure 3, significant (see METHODS – Signal Processing for explanation of “significant”) beta peak was found in at least one side per each patient. More specifically, the highest beta band activity was found in patients 01, 02, and 03 in contact pair 0–2 (left side), 0–3 (left side), and 10–11 (right side), respectively.

From the internal memory of the IPG, we downloaded the PSD of the contact pair selected from streamed LFPs, thus allowing a full monitoring of LFP fluctuation over time. The AlphaDBSipg stored the power spectrum every 10 min and downloaded it (together with the beta power value every minute and the stimulation value every 10 min) on the AlphaDBSpat at each recharging. Figure 4 shows the time-frequency spectra during 6 h of chronic recordings in the short-term follow-up conducted in clinic for each of the three patients. As shown, the embedded mode provided high-quality data that were successfully used for closed-loop stimulation. Note that, for patients 01 and 02, the recording configuration was symmetric (stimulation contact between recording contacts), whereas for patient 03, it was asymmetric (stimulation contact outside recording contacts). Iterative storing and downloading allowed for chronic data collection, providing highlights on the system functioning in terms of beta power tracking, closed-loop implementation, and neurophysiological monitoring.

DBS parameter settings for all patients in cDBS mode are reported in Table 2. Despite the small number of patients, which prevents from running a statistical comparison, stimulation parameters were similar, for cDBS, using the AlphaDBSipg and the previously implanted IPG. The total electrical energy delivered per second (TEED) by AlphaDBSipg in cDBS was lower than the previous IPG in patient 01 (−116 μW), in patient 02 (−55 μW), and higher in patient 03 (+ 29 μW) (see Table 2).