I-Ta Tsai MD, Presenter: Nothing to Disclose

Meng-Yuan Tsai, Abstract Co-Author: Nothing to Disclose

Ming-Ting Wu MD, Abstract Co-Author: Nothing to Disclose

Kuen-Huang Chen MD, Abstract Co-Author: Nothing to Disclose

We constructed an efficient and reliable AR web application of BMD. It prevented human errors from technicians, typists and radiologists. It facilitates current clinical service and has potential academic applications.

The reports of bone mineral density (BMD) are based on a diagnostic algorithm on the numeric data. In retyping these data manually, a few mistakes might be made, and the process is time-consuming. We constructed an web application using Ruby on Rails, an open source web application framework. By importing the data generated by a dual-energy x-ray absorptiometry (DXA) scanner, the web application can automatically generate structure reports integrated with the electronic medical records.  

For comparison of reporting speed, in Jan 2014, 500 examinations were randomized into Automatic Group (AG) and Manual Group (MG). With 25 examinations per test unit, the average time spent of report generation in AG and in MG (dictation and check) was 264 seconds and 1,452 seconds, respectively (p < 0.001). For evaluation of the accuracy, 5,120 examinations during Jan 2013 and Dec 2013 were enrolled retrospectively. With an AutoHotKey script, the context of automatically generated reports (AR) were compared with the formal manual reports (MR). There were 383 discrepant reports. The accuracy of calculation of T and Z scores in AR is 100%. The errors in AR were key-in errors by technicians (0.64%, 33/5120) and need of additional judgements (0.57%, 29/5120); in MR, there were misreading of T or Z score (2.32%, 119/5120), mis-assignment of hip level (2.17%, 111/5120), dictation error (1.21%, 62/5120) and data omission (0.57%, 29/5120). The overall accuracy of AR and MR is 98.8% and 93.7%, respectively (P < 0.001). The mis-categorization of BMD in AR and MR is 0.039% (2/5120) and 0.273% (14/5120), respectively (P = 0.17).  

Ancillary radiological comments such as degenerative change of spine with relative high T scores in MR were not generated in current AR. Revised algorithm would give a warning message in the future AR. The structured database could be integrated for epidemiological statistics such as normal range of our population.  

Tsai, I, Tsai, M, Wu, M, Chen, K, An Automatic Bone Mineral Density Reporting System  .  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14003842.html