Dobie R A, Wilson M J
Department of Otolaryngology-Head and Neck Surgery, University of Texas Health Science Center at San Antonio, USA.
Ear Hear. 1995 Jun;16(3):299-310. doi: 10.1097/00003446-199506000-00007.
Objective threshold estimation is possible using statistics such as magnitude-squared coherence (MSC) and responses such as the 40-Hz auditory evoked potential (AEP). The purpose of this paper is to compare three general methods of threshold estimation using MSC.
Using 500-Hz amplitude-modulated tones and 40-Hz AEPs from 10 paid-volunteer young adults with normal hearing, we compared three MSC-based threshold estimation methods--extrapolation, intensity series, and adaptive.
The extrapolation method yielded low thresholds but unacceptable variability. The intensity series method was insensitive and time-consuming. Two adaptive methods displayed good performance in terms of precision and sensitivity. The first employed an MSC criterion with a 13.5% false positive rate (per trial), but achieved overall false positive rates under 5% after 5 to 7 runs, requiring 30 to 40 minutes test time (a run is a single set of ascending or descending trials, terminated by a reversal). The second, which performed better overall, included only a single run, but collected more data at intensities near threshold than at higher intensities; test time could be further reduced (from 22 to 18 minutes) by using a stopping rule based on initial MSC and noise power estimates at each test intensity. If speed is paramount, a simple single-run method achieved threshold estimates in about 9 minutes, starting at 40 dB nHL, with acceptable precision but less sensitivity.
A single-run adaptive method, with or without collection of additional data near threshold, yielded the best results, in terms of statistical performance and data collection time.
使用诸如幅度平方相干（MSC）等统计量以及诸如40赫兹听觉诱发电位（AEP）等反应来进行客观阈值估计是可行的。本文的目的是比较三种使用MSC进行阈值估计的一般方法。
我们使用来自10名听力正常的有偿志愿者青年成年人的500赫兹调幅音和40赫兹AEP，比较了三种基于MSC的阈值估计方法——外推法、强度序列法和自适应法。
外推法得出的阈值较低，但变异性不可接受。强度序列法不敏感且耗时。两种自适应方法在精度和灵敏度方面表现良好。第一种方法采用了误报率为13.5%（每次试验）的MSC标准，但在5至7次运行后总体误报率低于5%，测试时间需要30至40分钟（一次运行是一组单一的升序或降序试验，以反转结束）。第二种方法总体表现更好，只进行一次运行，但在接近阈值的强度下比在较高强度下收集了更多数据；通过使用基于每次测试强度的初始MSC和噪声功率估计的停止规则，测试时间可以进一步缩短（从22分钟缩短到18分钟）。如果速度至关重要，一种简单的单次运行方法从40 dB nHL开始，大约9分钟就能得出阈值估计，精度可接受但灵敏度较低。
就统计性能和数据收集时间而言，一种单次运行的自适应方法，无论是否在阈值附近收集额外数据，都能产生最佳结果。
