Individuals with autism spectrum disorder have altered visual encoding capacity

Abstract

Perceptual anomalies in individuals with autism spectrum disorder (ASD) have been attributed to an imbalance in weighting incoming sensory evidence with prior knowledge when interpreting sensory information. Here, we show that sensory encoding and how it adapts to changing stimulus statistics during feedback also characteristically differs between neurotypical and ASD groups. In a visual orientation estimation task, we extracted the accuracy of sensory encoding from psychophysical data by using an information theoretic measure. Initially, sensory representations in both groups reflected the statistics of visual orientations in natural scenes, but encoding capacity was overall lower in the ASD group. Exposure to an artificial (i.e., uniform) distribution of visual orientations coupled with performance feedback altered the sensory representations of the neurotypical group toward the novel experimental statistics, while also increasing their total encoding capacity. In contrast, neither total encoding capacity nor its allocation significantly changed in the ASD group. Across both groups, the degree of adaptation was correlated with participants' initial encoding capacity. These findings highlight substantial deficits in sensory encoding-independent from and potentially in addition to deficits in decoding-in individuals with ASD.

Fig 1. Theoretical and conceptual framework.

(A) Perception can be described as an encoding–decoding process. Stimulus is first encoded in the noisy and resource limited sensory representation m. An estimate $\widehat{\mathrm{\theta }}$ is generated given m based on the decoding process. As an experimenter, we can only measure overt responses; characterized as a participant’s bias ($\mathrm{\mu }\left(\widehat{\mathrm{\theta }}\right)-\mathrm{\theta }$; difference between the average estimate and real stimulus value) and variance σ²(θ) in their responses. (B) The fidelity of encoding in neurotypical participants is generally anisotropic due to uneven allocation of sensory resources determined by environmental statistics. For example, as shown here, the nonlinear transformation between stimulus values and a neural space (solid red line) with homogenous noise results in higher uncertainty for orientations around the oblique than those around the cardinal. This can be characterized by a nonuniform profile of Fisher information I_F(θ) (solid green line). In our experiment, we imposed an artificial, uniform distribution of orientations, different from that of natural environment, which allows us to study whether and how both groups update their encoding under changing stimulus statistics. (C) Cramer–Rao lower bound specifies the lawful relationship between bias b(θ), variance σ²(θ), and encoding accuracy, i.e., Fisher information I_F(θ), regardless of the decoding scheme. This allows us to directly characterize encoding, while remaining agnostic about the details of the decoding process (also see S1 Text). (D) To demonstrate validity of our approach, we simulate an observer with anisotropic encoding process, p(m|θ), with a peak FI at 90°. (E) As an example, we construct 3 arbitrary decoders (red, black, and blue), yielding very distinct pattern of estimation biases and variances, yet they all attain the Cramer–Rao lower bound. (F) Applying the inequality, we estimate the (lower bound of) FI, which appropriately recovers the identical, true underlying pattern of FI in the encoding regardless of idiosyncrasies in the decoding process. Raw data and code underlying this figure can be found at S1 Code, and numerical values that make up this figure can be found at S1 Data. FI, Fisher information.

Fig 2. Experimental protocol and individual participant performance.

(A) A target orientation (Gabor) is briefly presented, and participants report their percept by orienting a line indicator (white) via left or right button press. No feedback is given on the first block of trials but is in subsequent blocks by overlaying the target orientation and the participant’s response. (B) Target orientations (x-axis) are drawn from a uniform distribution (individual dots are single trials). Y-axis indicates the bias for an example neurotypical participant, and lines are the running average within a sliding window of 18°. Different columns (and the color gradient) respectively show performance on the block without feedback (woFB; leftmost), on the first block with feedback (wFB1; center), and the second block with feedback (wFB2; rightmost). (C) follows the format in (B) while depicting the performance of an example ASD participant. Raw data and code underlying this figure can be found at S1 Code, and numerical values that make up this figure can be found at S2 Data. ASD, autism spectrum disorder.

Fig 3. Orientation perception in combined neurotypical and ASD participant.

Bias (y-axis) as a function of target orientation (x-axis, cardinal and oblique orientation indicated by dashed lines) and feedback block in neurotypical (A) and ASD (B) participants. Variance (1/κ, y-axis, see Materials and methods) as a function of target orientation and feedback block in neurotypical (C) and ASD (D) participants. RMSE (y-axis) as a function of target orientation and feedback block in neurotypical (E) and ASD (F) participants. Smoothing using a sliding window of 18° has been applied for visualization purpose. Error bars are ± SEM across 5,000 bootstrap runs. Raw data and code underlying this figure can be found at S1 Code, and numerical values that make up this figure can be found at S3 Data. ASD, autism spectrum disorder; RMSE, root-mean-square error.

Fig 4. Quantification and parametrization of FI in neurotypical and ASD individuals.

(A) FI peaked at cardinal orientations for both neurotypical (blue) and ASD (red) individuals, similar to the natural scene statistics of orientations. Further, visual inspection suggests a flattening of this function with feedback in neurotypical participants (from dark to light blue) but not ASD participants (see panels C and E for quantification). Smoothing using a sliding window of 18° has been applied for visualization. (B) The total amount of FI was larger in neurotypical individuals than ASD at the outset and increased over the course of the experiment in neurotypical (blue color gradient) but not ASD (red color gradient) individuals. (C) The FI pattern as a function of orientation was quantified by 2 parameters, ω, which mixes a cardinal orientation prior with a uniform distribution as the normalized (square root of) FI, and λ, scaling total FI. (D) λ and (E) ω as a function of group (blue = neurotypical control, red = ASD) and block. Error bars are ± SEM across 5,000 bootstrap runs. Raw data and code underlying this figure can be found at S1 Code, and numerical values that make up this figure can be found at S4 Data. ASD, autism spectrum disorder; FI, Fisher information.

Fig 5. Correlation between sensory encoding capacity and flexibility in encoding resources allocation.

(A) The flatness of FI after feedback (y-axis, indicating incorporation of stimulus statistics from the experiment) correlates strongly with total encoding resources before feedback (x-axis). The blue dots indicate individual neurotypical participants, and red dots indicate individual ASD participants. (B) The same correlation holds within both the neurotypical and ASD groups. Raw data and code underlying this figure can be found at S1 Code, and numerical values that make up this figure can be found at S5 Data. ASD, autism spectrum disorder; FI, Fisher information.