Title:Automated Medical Device Display Reading Using Deep Learning Object Detection

Abstract:Telemedicine and mobile health applications, especially during the quarantine imposed by the covid-19 pandemic, led to an increase on the need of transferring health monitor readings from patients to specialists. Considering that most home medical devices use seven-segment displays, an automatic display reading algorithm should provide a more reliable tool for remote health care. This work proposes an end-to-end method for detection and reading seven-segment displays from medical devices based on deep learning object detection models. Two state of the art model families, EfficientDet and EfficientDet-lite, previously trained with the MS-COCO dataset, were fine-tuned on a dataset comprised by medical devices photos taken with mobile digital cameras, to simulate real case applications. Evaluation of the trained model show high efficiency, where all models achieved more than 98% of detection precision and more than 98% classification accuracy, with model EfficientDet-lite1 showing 100% detection precision and 100% correct digit classification for a test set of 104 images and 438 digits.