Audiogram Digitization Tool for Audiological Reports

Abstract

Multiple private and public insurers compensate workers whose hearing loss can be directly attributed to excessive exposure to noise in the workplace. The claim assessment process is typically lengthy and requires significant effort from human adjudicators who must interpret hand-recorded audiograms, often sent via fax or equivalent. In this work, we present a solution developed in partnership with the Workplace Safety Insurance Board of Ontario to streamline the adjudication process. We present a flexible and opensource audiogram digitization algorithm capable of automatically extracting the hearing thresholds from a scanned or faxed audiology report as a proof-of-concept. The algorithm extracts most thresholds within 5 dB accuracy, allowing to substantially lessen the time required to convert an audiogram into digital format in a semi-supervised fashion, and is a first step towards the automation of the adjudication process. The source code for the digitization algorithm and a desktop-based implementation of our NIHL annotation portal is publicly available on GitHub https://github.com/GreenCUBIC/AudiogramDigitization.

A Novel Method for Audiogram Digitization in Audiological Reports

“…In the context of audiograms, studies have integrated machine learning and image processing techniques for complete digitization, such as [11] and [12]. Li et al's Multi-stage Audiogram Interpretation Network (MAIN) [11] utilized multiple convolutional neural networks to extract audiograms, audiological symbols, and axis labels from audiogram reports.…”

“…In contrast, Chairh et al's method [12], which handled PTA symbols, was trained on a dataset nearly eight times larger than the Open Audiogram dataset. However, this method could not account for situations in which a participant did not respond at certain frequencies, as indicated by an arrow below the audiological symbol.…”

“…It also faced challenges with symbol recognition in cases of symbol overlap and required improvements in the accuracy of pixelto-frequency-threshold value mapping. [12] employed the traditional Hough Transform [13] for gridline detection in the audiogram. Subsequently, they used YOLOv5 [3] to identify axis labels and matched them with the detected lines to obtain the position information of each axis label in the image.…”

“…Similar to Chairh et al's work [12], we also use YOLOv5 [3] to identify labels and use Tesseract [4] to identify the labels. However, we proposed a novel workflow to address the audiograms.…”

“…However, we proposed a novel workflow to address the audiograms. Our proposed method comprehensively addresses these issues in [12]. It accommodates all PTA symbols, including the arrow symbol, as well as the sound field test that is often performed in conjunction with PTA.…”