The goal of this project was to support a still ongoing study of the Department of Neurology of the University of Würzburg dealing with the ultrasonographic investigation of peripheral nerves in the human body. For each subject 225 measurements are performed and stored as a pixel structure in the corresponding ultrasound image.
Collecting and typing out all results manually is a time-consuming, monotonous, and tiring process bearing a high risk of erroneous entries. Therefore, a lightweight tool was developed which reads a DICOM image, automatically extract the region of interest by using standard image processing methods, extracts the required information (OCR) and stores it into a structured Excel table.
Now the user has the opportunity to check the intermediate results for errors. This process is assisted by highlighting values which didn´t pass several preceding consistency checks.
After the reviewing phase the collected data is automatically transferred into the final data table.
During testing the tool achieved an excellent recognition rate of 99.94%. The human effort was dramatically reduced from an estimate of 55 minutes down to below four minutes per subject.