Automatic Measurement of Frequency and Decibel Information of Audiological Symbols in Handwritten Audiograms

Authors: Beyza Baştürk1, Fatma Nur Kömür2, Abbas Memiş3, Hasan Güçlü1

1İstanbul Medeniyet Üniversitesi
2Marmara University
3İstanbul University

Background: The results of hearing tests are written on an audiogram paper in many audiology clinics. Audiograms of hearing test findings are generally stored physically or by transferring it to a computer through a scanning process. In recent years, there have been studies on storing archived data by digitally transferring it to computers. However, there is a strong need for artificial intelligence model-based softwares that can detect audiological symbols frequently used in audiology clinics to automatically digitize audiogram papers.

Objective: The aim of this study is to develop an artificial intelligence model for the digital transfer of audiograms to computers. It is aimed to accurately calculate the frequency and decibel values of the symbols detected on the audiogram with the software and to analyze the accuracy of the calculations.

Methods: The detection of audiogram symbols was performed by using the YOLO object detector. In position extraction process of the symbols, Houghline transform was employed and a coordinate system is constructed by using the Hough lines. Then, the frequency and decibel information of the detected audiological symbols were extracted based on their positions on the audiogram coordinate system. The results were visualized using Python libraries and reported with the software developed.

Results: In the experimental analysis, performance tests were performed on 100 audiograms. According to the qualitative observations of two independent speacilist audiologists, the frequency and decibel information of audiological symbols were calculated correctly at a rate of 76%. For the symbols marked by hearing response, a success rate of 75% was achieved.

Conclusion: This study enabled the successful identification of symbols in audiograms using an artificial intelligence model. Therefore, it is thought that large audiogram sets can be digitized both rapidly and successfully.