Bayesian quantification of sensory reweighting in a familial bilateral vestibular disorder (DFNA9)

Abstract

DFNA9 is a rare progressive autosomal dominantly inherited vestibulo-cochlear disorder, resulting in a homogeneous group of patients with hearing impairment and bilateral vestibular function loss. These patients suffer from a deteriorated sense of spatial orientation, leading to balance problems in darkness, especially on irregular surfaces. Both behavioral and functional imaging studies suggest that the remaining sensory cues could compensate for the loss of vestibular information. A thorough model-based quantification of this reweighting in individual patients is, however, missing. Here we psychometrically examined the individual patient's sensory reweighting of these cues after complete vestibular loss. We asked a group of DFNA9 patients and healthy control subjects to judge the orientation (clockwise or counterclockwise relative to gravity) of a rod presented within an oriented square frame (rod-in-frame task) in three different head-on-body tilt conditions. Our results show a cyclical frame-induced bias in perceived gravity direction across a 90° range of frame orientations. The magnitude of this bias was significantly increased in the patients compared with the healthy control subjects. Response variability, which increased with head-on-body tilt, was also larger for the patients. Reverse engineering of the underlying signal properties, using Bayesian inference principles, suggests a reweighting of sensory signals, with an increase in visual weight of 20-40% in the patients. Our approach of combining psychophysics and Bayesian reverse engineering is the first to quantify the weights associated with the different sensory modalities at an individual patient level, which could make it possible to develop personal rehabilitation programs based on the patient's sensory weight distribution. NEW & NOTEWORTHY It has been suggested that patients with vestibular deficits can compensate for this loss by increasing reliance on other sensory cues, although an actual quantification of this reweighting is lacking. We combine experimental psychophysics with a reverse engineering approach based on Bayesian inference principles to quantify sensory reweighting in individual vestibular patients. We discuss the suitability of this approach for developing personal rehabilitation programs based on the patient's sensory weight distribution.

Keywords: bilateral vestibular areflexia; internal models; multisensory integration; rod-frame illusion; spatial orientation; verticality perception.

Fig. 1.

A: experimental procedure of the rod-and-frame task. After presentation of a square frame for 250 ms, a rod is briefly (33 ms) flashed within the frame. When the rod disappears, the square remains visible until the subject responds as to whether the rod was rotated CW or CCW from upright. A 500-ms black screen is presented before the start of a new trial. B: schematic representation of the Bayesian model for verticality perception. For an optimal estimate of head-in-space orientation, used in the rod-and-frame task, the model integrates the global visual context (θ_R) together with vestibular/nonvestibular information (H_S) and prior information (H_P) that the head is usually upright. Sensory signals are assumed to be accurate but contaminated with noise (visual: σ_ver, σ_hor; vestibular/nonvestibular: α_HS, β_HS; prior: σ_HP). The perceived orientation of the rod in space is then obtained by coordinate transforming the optimal estimate of head-in-space orientation with the eye-in-head orientation (uncompensated ocular counterroll, A_OCR) and the line-on-eye orientation (assumed to be veridical).

Fig. 2.

A: probability of CW responses plotted against rod orientation for 3 exemplar frame orientations (20° CCW, upright, and 20° CW) in a representative bilateral vestibular areflexic patient (filled circles) and an age- and sex-matched healthy control subject (open circles). In each panel, the solid lines plotted through the data represent the psychometric functions, quantifying the bias (µ, dashed line) and response variability of the subject. B and C: bias and response variability with frame orientation for the representative patient and control subject in A for all 3 head-in-space orientations. Solid lines plotted through the data represent the best fit of the Bayesian optimal integration model, fitted to all responses simultaneously. Best-fit parameters are found in Table 2 (participants P15, C15).

Fig. 3.

Mean bias (A) and response variability (B) plots across all patients (n = 16) and control subjects (n = 16). Error bars on the data represent the standard deviation across subjects. Solid lines through the data show the mean best fit across all patients and control subjects, with the shaded areas indicating the standard error on the model best fits.

Fig. 4.

Mean vertical (σ_ver) and horizontal (σ_hor) visual context noise across subjects (patients, n = 16; control subjects, n = 16) plotted against frame orientation. Shaded areas represent the standard error on the noise parameter values.

Fig. 5.

A: mean vertical visual context, nonvestibular/vestibular, and prior weight distributions over frame orientation in all 3 head-in-space orientations for patients (n = 16, top) and healthy control subjects (n = 16, bottom). Shaded areas represent the standard error over the mean weights. B: distribution of mean vertical visual context, vestibular/nonvestibular, and prior sensory weights in patients (left) and control subjects (right) at maximum vertical visual weight. Error bars represent the standard error on the distribution of weights across subjects.

Fig. 6.

Individual vertical visual context weights plotted against nonvestibular/vestibular weights in patients (n = 16; left, filled circles) and control subjects (n = 16; right, open circles) for an upright frame and an upright seated subject. Crosses represent the mean weights in the upright condition of Fig. 5B.