# Automated Medical Device Display Reading Using Deep Learning Object Detection

**Authors:** Lucas P. Moreira  
**arXiv:** [2210.01325](https://arxiv.org/abs/2210.01325)  
**Date:** 2022-10-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.01325v1/

## Key points

- COVID-19 pandemic has increased the need for remote health care through telemedicine and mobile health applications.
- Automated tools are in demand to transfer health monitor readings from patients to specialists.
- Most home medical devices use seven-segment displays, which can be difficult to read accurately without human intervention.
- Researchers have proposed an end-to-end method for detecting and reading seven-segment displays from medical devices using deep learning object detection models.
- The study used two state-of-the-art model families, EfficientDet and EfficientDet-lite, which were previously trained with the MS-COCO dataset.
- The models were fine-tuned on a dataset of medical device photos taken with mobile digital cameras to simulate real-world case applications.
- The evaluation of the trained models showed high efficiency, with all models achieving more than 98% detection precision and more than 98% classification accuracy.
- Automatic display reading algorithms based on deep learning object detection models could provide reliable tools for remote health care.
- This study adds to the growing body of research on deep learning based solutions in various fields such as smart hats for the blind with real time object detection using Raspberry Pi and TensorFlow Lite; design optimization of TensorFlow Lite deep learning neural networks for human activity recognition on smartphones; efficient ship target detection of unmanned surface vehicles based on efficientdet; fabric defect detection based on edge computing using efficientdet; diabetic foot ulcer detection using deep learning; and waste detection in natural and urban environments using deep learning based methods.

## Summary

The COVID-19 pandemic has led to an increase in the need for remote health care, particularly through telemedicine and mobile health applications. As a result, there is a growing demand for automated tools that can transfer health monitor readings from patients to specialists. Most home medical devices use seven-segment displays, which can be difficult to read accurately without human intervention. To address this issue, researchers have proposed an end-to-end method for detecting and reading seven-segment displays from medical devices using deep learning object detection models. The study used two state-of-the-art model families, EfficientDet and EfficientDet-lite, which were previously trained with the MS-COCO dataset. The models were fine-tuned on a dataset of medical device photos taken with mobile digital cameras to simulate real-world case applications. The evaluation of the trained models showed high efficiency, with all models achieving more than 98% detection precision and more than 98% classification accuracy. In particular, model EfficientDet-lite1 achieved 100% detection precision and 100% correct digit classification for a test set of 104 images and 438 digits. These results suggest that automatic display reading algorithms based on deep learning object detection models could provide reliable tools for remote health care. This study adds to the growing body of research on deep learning based solutions in various fields such as smart hats for the blind with real time object detection using Raspberry Pi and TensorFlow Lite; design optimization of TensorFlow Lite deep learning neural networks for human activity recognition on smartphones; efficient ship target detection of unmanned surface vehicles based on efficientdet; fabric defect detection based on edge computing using efficientdet; diabetic foot ulcer detection using deep learning; and waste detection in natural and urban environments using deep learning based methods. Overall, this study provides valuable insights into how deep learning object detection models can be used to automate medical device display reading which could improve the accuracy and reliability of remote health care. The authors report no conflict of interests or specific grant funding from any public commercial or not for profit sectors.