Performance of an Automated Detection Algorithm to Assess Objective Pulsatile Tinnitus.

Abstract

In this paper we describe an automated detection algorithm that objectively detects pulsatile tinnitus (PT) and investigate its performance. Sound measurements were made with a sensitive microphone placed in the outer ear canal in 36 PT-patients referred to our tertiary clinic, along with a registration of the heart rate. A novel algorithm expressed the coherence between the recorded sound and heart rate as a pulsatility index. This index was determined for 6 octave bands of the recorded sound. We assessed the performance of the detection algorithm by comparing it with the judgement of 3 blinded observers. The algorithm showed good agreement compared with the majority judgement of the blinded observers (ROC AUC 0.83). Interobserver reliability for detecting PT in sound recordings by the three blinded observers was substantial (Fleiss's κ=0.64). The algorithm may be a reliable alternative to subjective assessments of in-canal sound measurements in PT-patients, thus providing clinicians with an objective measure to differentiate between subjective and objective pulsatile tinnitus.