Individual radiologists may not always go about interpreting an imaging study in the same manner on PACS as their colleagues, but these variations don't typically result in differences in reading times, according to research presented at the recent RSNA 2015 meeting in Chicago.
Employing a form of analytics called process mining, a team of researchers from PACS vendor Sectra and Case Western Reserve University in the U.S. studied PACS usage logs and found that radiologists who had a more complex workflow for reading exams were just as fast as those who followed a simple interpretation process. They also found that radiologists tended to follow the same patterns over time for interacting with the PACS software.
The researchers believe that these and future insights derived from their process mining technique may ultimately help to develop "best practices" for reading imaging studies on PACS.
"At the moment, the input data we use is not only useful for process mining but can also be utilized in PACS usage analytics, where we analyze the data on both a descriptive and diagnostic level in order to improve PACS utilization," said presenter Daniel Forsberg, PhD, a research scientist at Sectra.
Efficiency gains
While many research studies have examined radiology workflow in hopes of improving efficiency, the Sectra and Case Western research team sought to zero in on the interpretation step, specifically focusing on the work performed by radiologists on the PACS workstation while they read imaging studies. They wanted to determine if there were any differences in how radiologists use PACS and if it was possible to identify any recurring usage patterns, Forsberg said.
The researchers wanted to apply process mining, an analytics method that extracts underlying process models from recorded event logs and analyzes them for any possible workflow improvements.
"In our case, we're interested in using process mining as a way to model the interaction patterns of radiologists and to use these models to identify workflow aspects related to PACS usage where improvements can be made, either by improved system/user configuration or by training of the users," Forsberg told AuntMinnieEurope.com.
The researchers studied usage data logs on commercially available PACS software (Sectra) for 567 single-view chest exams that had been reviewed by 14 radiologists over one week at Case Western. After recording specific factors for each case (the radiologist, specialty, and time of day), they then performed statistical analysis to see if these factors correlated with three variables (the number of commands made by the radiologist to the PACS software, the number of command classes, and the time to read a case). Next, they applied process mining to discover any potential underlying PACS usage patterns, he said.
Surprisingly, they found only a slightly positive correlation between the number of commands and the number of command classes used per case with the time to read a study, Forsberg said. In other findings, the group discovered that the specific radiologist, specialty, and time of day all had an effect on the number of commands used during interpretation. The specific radiologist also affected the number of command classes used in the reading process. None of these had a statistically significant impact on the time to read a study, however, Forsberg said.
A spaghetti-like process model
After then applying process mining techniques to the PACS usage logs from all of the radiologists, the researchers found a "spaghetti-like" process model that demonstrated no obvious overall PACS usage pattern. However, it was possible to discern frequently used commands as well as command combinations, Forsberg said.
More obvious PACS usage patterns emerged, however, when looking specifically at individual radiologists. There were large differences in terms of complexity between the discovered process models, but again, there were no significant differences in the time it took to interpret a study, he said.
The majority of the radiologists in the study followed a structured process for interacting with the PACS, while a smaller group had a more complex usage pattern and used lots of different commands, he said.
To see if the radiologists were consistent in how they used PACS, the researchers performed the same analysis on the same radiologists for the following week. They again observed a "spaghetti-like" process model when considering all radiologists; frequently used commands and command combinations also were roughly the same, Forsberg said.
PACS usage patterns differ among radiologists for interpreting chest radiographs, and process mining was useful for deriving PACS usage patterns, the researchers concluded.
"Individual patterns appear stable over time," he said.
In the future, the researchers plan to extend their data to include more exam types and longer time periods. They also would like to add performance analysis, Forsberg said.
"The end game is gaining better understanding in terms of how [radiologists] interact with PACS and [exploring] the possibility of deriving best practices for different types of examinations," he said.
