Nibeditaa SpeaksA new automated method uses optical character recognition and a voting algorithm to extract numerical hearing-test results from medical report images, potentially making audiological research more efficient.
Hearing-test records locked inside medical report images could become easier to analyse with an automated data-extraction method designed to reduce the time and effort required for manual transcription.
The method focuses on extracting numerical data from audiometric reports stored as images in electronic medical record (EMR) systems. It combines optical character recognition (OCR) with a voting-based algorithm to improve the reliability of information captured from these documents.
The development addresses a practical challenge in hearing healthcare: although audiometric tests are routinely used to assess patients’ hearing, their results are not always available in formats that researchers can readily analyse. Medical institutions may store the information in tabular reports or plotted audiograms, requiring researchers to manually transfer results into usable datasets.
The process can be labour-intensive, time-consuming and costly, while manual data entry can introduce errors.
Why digitising hearing-test records matters
Hearing loss affects more than 1.5 billion people worldwide, according to the World Health Organization’s 2021 World Report on Hearing. At least 430 million people require professional care for disabling hearing loss.
Untreated hearing impairment can affect communication, language development, education, mental health and economic independence. Reliable hearing-test data are therefore important for clinical assessment, treatment planning and research.
Audiometric evaluations commonly include pure-tone audiometry (PTA), speech reception threshold (SRT) and word recognition score (WRS). These measurements help clinicians assess different aspects of a person’s hearing ability.
However, when the results remain embedded in image files, researchers often have to extract and enter the information manually before conducting large-scale analysis.
How the automated extraction method works
The proposed system uses Tesseract, an open-source OCR engine, to recognise text and numbers in audiometric report images. It incorporates a voting optical character recognition (VOCR) algorithm, which combines multiple OCR outputs to improve the reliability of extracted information.
Image preprocessing is used to improve recognition before the system processes the report. The voting mechanism then compares the outputs and provides reliability indicators, helping data collectors identify entries that may require manual review.
Instead of checking every extracted value individually, researchers can prioritise sections where the recognition results show less agreement. This could help streamline data collection while retaining a human review process for uncertain entries.
The method also addresses special characters in SRT and WRS readings that OCR systems may misinterpret. Removing these symbols can improve the recognition of numerical information.
Addressing a gap in existing technology
Earlier approaches to automated audiometric data extraction have largely focused on graphical audiograms, where hearing thresholds are represented by symbols and curves plotted against frequency and intensity.
Researchers have used computer vision, deep-learning models and OCR to identify these plotted symbols and convert them into numerical measurements.
However, many medical and occupational-health settings also use tabular reports containing numerical hearing thresholds. These tables present a different challenge from graphical audiograms and may still require manual transcription.
By targeting table-formatted reports, the proposed system offers a complementary approach to existing tools. Its developers aim to support the extraction of data from different report formats, potentially facilitating research involving multiple medical institutions.
Potential benefits for healthcare research
Automating the extraction of audiometric records could reduce the workload associated with preparing datasets for clinical analysis and research. It may also help minimise transcription errors and make previously difficult-to-use records more accessible for large-scale studies.
The method is intended to support more efficient data collection rather than replace clinical judgement. Its reliability indicators allow uncertain results to be flagged for review instead of treating every automated reading as unquestionably correct.
The approach could help researchers make better use of existing hearing-test records, although its practical value will depend on recognition performance across different report formats and clinical settings.
As healthcare research increasingly relies on large datasets, tools that convert medical documents into structured information could help bridge the gap between routinely collected clinical records and the data needed for analysis.
Add comment