Deep Learning Based Automatic Detection of Audiological Symbols in Audiogram Images

Hard copy documents in clinical archives need to be digitized and registered to the digital health systems in order to be assessed and tracked through these systems. As it is known, thanks to the machine learning and image processing-based approaches, basic tasks such as automatic detection, segmentation and classification can be performed on a large number of handwritten documents very quickly and these documents can be digitized. In this paper, a research study on the automatic detection and digitization of various audiological symbols in audiogram images is presented. In the proposed study, it is aimed to detect automatically 19 different audiological symbols that are likely to appear on audiogram images by using deep learning-based object detection methods. By using the YOLO (You Only Look Once) object detection algorithm, quite successful results were observed in the experimental studies performed on 1000 audiogram images, and a value of 0.9600 precision, 0.9700 recall, 0.9650 F1-score, 0.9580 mAP 50 and 0.6730 mAP50-95 were measured for all 19 different audiological symbols. The proposed study is one of the limited studies in the current literature on the automatic detection of audiological symbols from the audiogram images, and it is also the first national study, to the best of our knowledge, on this topic.