House and Hitselberger reported in 1979 the first results of an auditory brainstem implant (ABI). They placed the ABI electrode via translabyrinthine craniotomy at the time of the vestibular schwannoma removal. They developed this device because the loss of hearing of their patient resulted from destruction of the cochlear nerves (bilateral vestibular schwannoma), rendering cochlear implantation ineffective. The cause of bilateral destruction of the cochlear nerves is mainly the neurofibromatosis type 2 (NF2). Their patient had benefit from direct stimulation of the cochlear nucleus and had used her device during all of her life (House, 2001).
This result has been confirmed by different teams with different devices from Cochlear Corporation, Medel, and Neurelec (Brackman, 1993; Shannon, 1993; Sollman, 2000; Vincent, 2005) and nowadays ABI are frequently used for NF2 patients. Cochlear, Medel and Neurelec devices have been described respectively by Sollman (2000), Behr (2007), and Vincent (2005). The main differences between these devices are the number of active channels and the speech processing strategy.
Worldwide, more than 500 persons have received an ABI after removal of the tumors that occur with NF2 (McCreery, 2008). Most patients with the implant have good appreciation of environmental sounds, but obtain more modest benefit with regard to speech perception with only the sound from the ABI. The device is most frequently used to facilitate lip reading.
More recently, some teams have proposed the use of ABI in patients without any tumors (bilateral total ossified cochlea, inner ear malformations, vestibular schwannoma with controlateral lesions) (Grayeli, 2003) or cochlear nerve aplasia (Coletti, 2005a).
A Dedicated Electrode
ABI are proposed by Cochlear corporation, Medel, and Neurelec. In all cases, ABI share a common concept with cochlear implants (CI).
Schematically (Fig. 1), the microphone picks up sound and sends it to the signal processor. The sound signal is processed in the external processor (adaptation of the dynamic range, tonotopic reorganization regarding to the number of active channels). The processor selects the particular characteristics of the sound (i.e., the speech envelope) and electronically encodes them. At low rate of stimulation (<500 Hz), the processor selects the number and location of the electrodes to be stimulated depending on the intensity and frequency of the incoming signal. At higher rate of stimulation (>500 Hz), the speech envelope is encoded by a continuous stimulation of a small number of channels. After this processing, the transmitter sends the signal to the internal receiver/stimulator. Finally, the electrical signal is sent by the receiver/stimulator directly to the ABI electrode array, and the electric pulses are delivered via plate electrodes placed at the surface of the cochlear nucleus.
The device comprises an external part similar to a cochlear implant device and an implanted portion with a specially designed electrode. This electrode is a pad designed for surface plate electrodes which is placed on the brainstem surface to stimulate directly the cochlear nuclei. There are differences in the number of channels between the different companies but all the pads have roughly the same dimensions adapted to the 3 by 8 mm surface of the CN (Fig. 2).
Penetrating electrodes for ABI (PABI) have been developed to reach full access of the tonotopy of the cochlear nucleus (CN) because a lot of patients with surface electrodes cannot pitch-range their electrodes. Typically with PABI, the threshold for auditory percept is low, ranging from 0.8 to 2.0 nC/ph because the electrodes are not separated from the CN by the glia-pia membrane. By comparison, the thresholds for auditory percepts with surface electrodes are much higher from 10 to 100 nC/ph (Mc Creery, 2008). Even if a better tonotopic access have been reported with PABI, there is still a relatively poor coverage of the low frequencies with these penetrating electrodes.
The auditory midbrain implant (AMI) is a new hearing prosthesis designed for stimulation of the inferior colliculus (Lim, 2009). The authors have begun clinical trials in which five patients have been implanted with a single shank AMI array (20 electrodes). Stimulation of the auditory midbrain provides a wide range of level, spectral, and temporal cues, all of which are important for speech understanding, but they do not appear to sufficiently fuse together to enable open set speech perception with the currently used stimulation strategies.
A Specific Target
The CN is located on the dorsolateral surface of the brainstem at the junction of the medulla and pons. In humans, the CN is rotated outward, forming an angle of 30°–35° between the rostrocaudal axis and the neuroaxis (Moore, 1979). It is a complicated heteromorphic formation consisting of different subunits, namely, the dorsal (DCN) and ventral (VCN) nuclei (Fig. 3). The latter comprises smaller subdivisions: the anteroventral cochlear nucleus (AVCN) lying superiorly and the posteroventral cochlear nucleus (PVCN) lying inferiorly (Fig. 4b) (Terr, 1985).
Each subunit has its own tonotopy and the population of auditory neurons is variable among the CN (Dublin, 1982). However, in humans, the topographical separation in the various cell types seems less complete than in other mammals (Moore, 1979). The CN contains at least six major types of cells characterized by specific properties (morphology, regional distribution, and cell-membrane characteristics), synaptic input and responses to acoustic stimuli (Figs. 4a and 5) (Cant, 1992).
A widely used approach to the classification of CN units is based on the shape of the post-stimulus-time histogram in response to short tone bursts (Kiang, 1965). Responses to tone and noise stimuli have been studied in the cat (Pfeiffer, 1966; Van Gisbergen, 1975).
Pfeiffer (1966) described four main patterns in response to tone stimuli: primary-like, chopper, on and pause. Van Giesbergen (1975) studied the frequency selectivity and quality of phase locking of these neurons. There was a strong variation in phase locking among neurons but phase locking could not be demonstrated for characteristic frequencies exceeding 4.5 kHz. Primary-like neurons recorded from spherical bushy cells located in the anterior part of the AVCN show precise phase locking to pure tone similar to that of acoustic nerve fibers. Their response is described as “primary-like” since it preserves quite well the structure and overall timing of the action potentials of the auditory nerve. Globular bushy cells located in the posterior part of the AVCN and anterior part of the PVCN response with a modification of the “primary-like” pattern called a “primary-like-with-notch” pattern, which is characterized by a high onset peak, followed by a short notch of little to no activity for 0.5–2 ms.
Chopper units recorded from stellate cells, located throughout the CN but more concentrated in the posterior part of the AVCN and the PVCN, poorly phase lock to pure tone above 1 kHz. Evidences for the presence of both excitatory and inhibitory inputs were found to be associated with poor phase locking capabilities. In contrast with the irregular auditory nerve firing pattern, chopping patterns feature high discharge regularity. However, the period of these peaks is not related to the period of the tone stimulus (Rhode, 2010). Chopper units can be further categorized as transient choppers, which fire regularly only at the beginning of the tone, and sustained choppers, which fire regularly throughout the duration of the tone burst.
Fusiform cells, predominantly found in the DCN are associated with “pause” pattern of response to tone burst. Anesthesia strongly influences the response patterns of DCN cells while there is no such evidence in the VCN (May, 1998).
Octopus cells are found in the caudal region of the PVCN. These cells are usually associated with an “on” pattern in response to tone burst, characterized by a short latency, sharply timed response at the beginning of the stimulus followed by little to no activity throughout the remainder of the tone burst.
To investigate how these different cell types in the CN respond to speech sounds, the mode of encoding of steady-state vowels have been studied (Palmer, 1986). The most studied unit type has been the primary-like unit recorded from bushy cells. These neurons provide a good representation of vowel formant in their fine temporal pattern of discharge. Therefore, this cell population is a good candidate for electrical stimulation with speech processing derived from cochlear implant. Chopper units with poor phase locking above 1 kHz, provide information about the first formant frequency but not about higher formants.
The consequence of this specific cytoarchitecture is that it is difficult to access to the tonotopy via a surface electrode placed against the DCN and the PVCN. Moreover, it is not known how electric pulses delivered by the surfaces electrodes of the ABI may mimic the different electrical pattern of responses of the different auditory neurons cell types (Fig. 5).
The DCN and PVCN are the target of stimulation. The DCN is the major target of stimulation since the PVCN may vary in anatomical situation (depth). In addition, the DCN is bigger, accounting for a volume varying from a half and a third that of the VCN (Moore, 1979). These structures are visible in the floor of the lateral recess of the fourth ventricle curving around the inferior cerebellar peduncle (Fig. 3). The size available for auditory stimulation is estimated to measure 3–8 mm long (Lejeune 1997, McElveen 1986). The surgical landmarks to access to that region are the stump of the eighth cranial nerve, the glossopharyngeal nerve and the choroid plexus. They usually lead to the entrance of the lateral recess of the IV ventricle as shown in Fig. 6. Location of the lateral recess can be confirmed by noting the egress of cerebrospinal fluid. However, these landmarks may be difficult to recognize in case of a large tumor with brainstem deformation. In these difficult cases, intraoperative electrophysiology (electrical early auditory brainstem responses) may help in refining the placement of the electrode within the lateral recess of the IV ventricle. Intraoperative testing can also detect stimulation of adjacent cranial nerves or changes in vital signs.
An Adapted Fitting
The first fitting is often performed 6 to 8 weeks after implantation. Some teams performed it with available emergency medical assistance in case of any cardiovascular side effects. However, intraoperative testing has been usually done and the electrical stimulation is performed starting from no current with small increase to reach threshold level.
As described, the target of stimulation is surrounded by other cranial nerves and brainstem nuclei and the surgical landmarks. Consequently, the electrical stimulation may elicit nonauditory sensations: dizziness, sensation of tingling in the leg, in the tongue. The first task of stimulation is to deactivate individual electrodes because of nonauditory responses.
The next step of the fitting procedure is to perform pitch scaling and ranking for electrodes that elicit auditory responses. Due to the tonotopic organization of the CN in subunits and layers and the surface stimulation, there is no definite tonotopic relationship between the ABI electrode and the CN. This is the reason why ABI with penetrating electrodes (PABI) have been developed to have access to a better tonotopy. Otto et al. (2008) reported lower threshold, increased pitch range, and high selectivity with PABI, but these properties did not result in improved speech recognition.
To define the appropriate tonotopic order of the electrodes for the sound processor, electrodes are stimulated at C-level, and the patient is asked to perform a pitch-ranking of the electrodes. Pitch-ranking is usually performed once the Threshold and Comfort levels, namely T- and C-levels, have been established. This defined tonotopic order is checked by sweeping all the active electrodes for adequate speech processing.
Speech processing varies amongst the companies. None of the CI companies involved in ABI technology have developed specific strategies for ABI for the moment. The best parameters for optimal CN stimulation (rate of stimulation, emphasis of spectral, or temporal characteristics) are not well defined as the optimal number of active channels.
Patients undergoing ABIs are followed carefully because changes in perceptual responses to stimulation may appear over time (Otto, 2002). Most of ABI recipients are NF2 patients. They can detect and discriminate sounds based on their temporal and amplitude properties but most of them cannot identify words or sentences with only the sound from the ABI. The benefit in communication is mainly obtained by the lip-reading improvement. In addition, many of these patients are able to recognize environmental sounds. Generally, in all publications, there is a lot of individual variability which could not be expressed adequately by classical statistics.
Some authors have reported better outcomes when ABIs are provided to patients who have lost their auditory nerve from causes other than NF2. Some patients have sentence understanding of more than 50% in sound-only mode (Colletti, 2005b). These results suggest that the cause of poor speech recognition in NF2 ABI patients may be related to NF2. However, first, these results have not been confirmed by all the ABI teams, and secondly, recent results in Europe have demonstrated speech recognition levels similar to those of CI patients in NF2 ABI patients (Behr, 2007). It is thought that a large number of channels increases auditory outcomes but these results were obtained with the use of the Medel ABI which has a 12-channel electrode (less than the Cochlear and Neurelec electrodes). Moreover, others have also reported good perceptual performance in patients with a low number of active channels. However, a minimum number of electrodes, four or more, would be necessary for adequate speech perception if we compare with our experience with cochlear implants (Dorman, 1997). We need certainly to know better about the functioning of the electrical stimulation at the CN level to isolate prognostic factors and to develop specific sound processing for CN stimulation. We know that most of the acoustic cues of the speech is delivered by amplitude coding with actual cochlear implant speech processing. This strategy may not be optimal for CN stimulation since place coding is difficult with surface electrodes and since the neurons stimulated do not share all the same pattern of response to sound. It could be interesting to test if fine structure could help in achieving better speech recognition performances. All CN neurons have phase-locking capabilities but not in the same range of frequency boundaries. The results reported by Behr (2007) could be the consequence of higher pulse rate of stimulation.
Electrical stimulation of the CN is a challenge due to the presence of different auditory neurons with different patterns of response (inhibitory, excitatory, chopper responses, etc). Access to the tonotopy of the CN with surface electrodes is difficult because the CN is organized in a three-dimensional structure and in subunits. However, initial results with PABI are comparable with those achieved by surface stimulation. To date, no specific speech processing strategies have been developed for ABI and we rely on the brain plasticity to achieve speech recognition with a “brute” strategy imported from our cochlear implant experience.
We do not how ABIs work but they work...