Stone Talie, Clark Torin K, Temple David R
University of Colorado Boulder (Molecular, Cellular, and Developmental Biology), Boulder, CO, USA.
University of Colorado Boulder (Smead Aerospace Engineering Sciences), Boulder, CO, USA.
Exp Brain Res. 2024 Dec 24;243(1):34. doi: 10.1007/s00221-024-06984-8.
In sensory perception, stochastic resonance (SR) refers to the application of noise to enhance information transfer, allowing for the sensing of lower-level stimuli. Previously, subjective-assessments identified SR in vestibular perceptual thresholds, assessed using a standard two alternative (i.e., binary), forced-choice task, when applying noisy Galvanic Vestibular Stimulation (nGVS). However, this required extensive testing of at least 100 binary trials to yield sufficiently precise thresholds at each of several nGVS amplitudes, leading to confounds of fatigue, sleepiness, learning, etc. stalling the study of vestibular SR. To mitigate this, we explore confidence reporting, which via a confidence signal detection (CSD) model may much more efficiently identify SR (i.e., with fewer trials), if SR exists in CSD thresholds. To test this, Y-translation thresholds were tested with 100 trials at each nGVS amplitude (0 or sham, 0.1, 0.2, 0.3 and 0.4 mA peak-to-peak). To objectively identify SR, we applied a machine learning classification algorithm trained on simulated datasets. We found significant evidence of SR exhibition using CSD thresholds (p = 0.0025), with six of 10 subjects classified as exhibiting SR. Next, we considered fewer trials, finding the false positive rate of SR identification to be better using CSD thresholds with as few as 50 trials, when compared to 100 binary trials. Applying the CSD model to our subject's data with a subset of their trials found similar classifications of SR exhibition as with 100 binary trials. We demonstrate CSD thresholds exhibit SR, proving a means of better and much more efficiently identifying SR.
在感官知觉中，随机共振（SR）是指通过施加噪声来增强信息传递，从而实现对较低水平刺激的感知。此前，主观评估在使用标准二选一（即二元）强制选择任务评估前庭知觉阈值时，通过应用噪声性电刺激前庭（nGVS）识别出了随机共振。然而，这需要进行至少100次二元试验的广泛测试，以便在几个nGVS振幅下分别得出足够精确的阈值，从而导致疲劳、困倦、学习等混淆因素阻碍了前庭随机共振的研究。为了缓解这一问题，我们探索了置信度报告，即如果在置信度信号检测（CSD）阈值中存在随机共振，那么通过该模型可能会更有效地识别随机共振（即使用更少的试验次数）。为了对此进行测试，在每个nGVS振幅（0或假刺激、0.1、0.2、0.3和0.4毫安峰峰值）下进行了100次试验来测试Y轴平移阈值。为了客观地识别随机共振，我们应用了一种在模拟数据集上训练的机器学习分类算法。我们发现使用CSD阈值有显著的随机共振表现证据（p = 0.0025），10名受试者中有6名被归类为表现出随机共振。接下来，我们考虑了更少的试验次数，发现与100次二元试验相比，使用低至50次试验的CSD阈值时，随机共振识别的假阳性率更低。将CSD模型应用于我们受试者部分试验的数据时，发现与100次二元试验相比，随机共振表现的分类相似。我们证明了CSD阈值表现出随机共振，证明了一种更好且更有效地识别随机共振的方法。
