Prior Perfection

Nearly everything about digital radiology is better than film. Digital images acquired onto PACS are available virtually anywhere in the world; they can be read at any time of day or night, and concurrently viewed at any number of locations; they can be enhanced by image algorithms that improve visibility; and they are largely safe from loss.

But there is a glaring deficiency to digital radiology: comparison hanging. The automatic retrieval of patients’ prior studies for comparison purposes is still ineffective. Why? Because no one has successfully designed software that has the intelligence to determine relationships among radiologic exams.

Let's look at comparison hanging—how it worked then, and how it works now.

With hard copy film, comparison studies were retrieved from a patient's records by radiology file clerks. These clerks typically had no post-secondary education, but with some training they could make accurate decisions as to what prior studies were relevant to a current study.

Today, image studies are electronically stored. There is no file clerk to judge what prior studies to retrieve for comparison purposes. It is the PACS software that does the retrieval.

Here is where the problems begin. Just as with computerized addressing of mail, a computer doesn't recognize similarity. Instead it wants exact matches. With a last name of Lidstrom, I might get four mailings from the same mail-order company, two with misspellings ("Lindstrom", for sure), and one using only the initial letter of my first name. That file clerk from 1980, though, would immediately recognize by context that they are all the same person.

And so it is with radiologic procedures. Someone has to "instruct" the PACS server that there may be several ways of naming even the most common procedures. PACS continues to suffer from the fact that there is no universal naming standard, and that study descriptions can be free-texted.

Consider how a single, common chest procedure, a 2-View Chest X-ray, might be described:

To a computer, these five exams are different. And the file clerk? A split-second glance and he knows that the study is the same; only the description is not.

Many believe that this problem will ultimately be resolved by using pre-defined study descriptions. When all procedures are selected from a standardized list, there can be no free-text variability. But it is not just an identical study that must compare to a current study. If our current study is a 2-view plain-film chest, all plain-film chest studies should hang as relevant priors, including these:

Many PACS companies have solved this problem . . . after a fashion. Their clients can have a spreadsheet or table that keeps track of related procedures. Before the PACS displays a prior study at the radiologist's workstation, it determines from the spreadsheet what procedures are related to the current study, compares that list against the list of studies in that patient's history, and displays the most recent prior.

But there are further complications. Many institutions have new or referred patients who bring studies from outside sources. These images, typically on CD, are often entered into the PACS for ease of viewing alongside current studies. The names of these studies almost certainly will deviate from the local institution’s. After a period of time we may have, in addition to the already cited chest procedures, many like these:

If this weren’t complication enough—and keep in mind that free-texted procedure descriptions exist for all the modalities and regions of the body—there is an even greater challenge that PACS software faces in reliably hanging comparison studies: cross-sectional imaging studies.

Consider a CT of the abdomen and pelvis. What would a radiologist want to see as a related prior exam? Any CT study that contains an abdomen or pelvis. This includes CTs of the abdomen only, the pelvis only, and cross-sectional studies of the Chest/Abdomen and the Chest/Abdomen/Pelvis.

Here are one institution’s CT exams from its RIS that would relate to a CT Abdomen/Pelvis.

Now consider free-texted Abdomen/Pelvis studies that entered this institution’s PACS from outside CDs. A query of their studies stored in PACS found over 500 uniquely-named exams that would relate to a single CT Abdomen/Pelvis. The related procedures for CT Abdomen/Pelvis alone are beyond the ability of a PACS administrator to manage.

This is the reality of today's PACS. The proliferation of thousands of procedure descriptions that enter the PACS from free-texting—whether generated locally or from outside sources—makes it impossible for PACS software to reliably hang prior exams, or PACS administrators to manually relate procedures. Pre-defined acquisition protocols, standardized naming conventions—even the RadLex committee’s creation of an anatomic naming schema—will not solve PACS’ ability to find relationships among radiologic procedures.

What's to be done?

One answer to this problem has been simply to accept that the PACS will automatically hang only the most common procedures. A plurality of procedures can hang appropriately with rigorous maintenance of a list of a few thousand related procedures. But is this the future of comparison hanging? Manual lists, and only a subset of patients' priors? What are the consequences of the PACS providing some, but not all, related priors?

At RadMapps, we have a better answer. We have created software that can deconstruct and categorize radiologic procedures by modality, “parallel” modality (e.g., CR and DR), body part functionality, sidedness, and body part proximity. Our work will allow cross-modality comparisons (e.g., MR and CT), find functional relationships among the myriad studies that exist throughout the PACS world, and create files that PACS companies can use to reliably relate radiologic image studies.

No longer will PACS administrators have to maintain lists of related procedures; no longer will only the most common procedures hang automatically. PACS companies can now be confident that their automated comparison hanging is thorough, accurate, and efficient. With RadMapps, your system will be able to search a patent's records, find historical studies that match the list of related procedures, and return all prior studies that are suitable for comparison-reading.

Voilà! What has eluded PACS providers for years will be possible: Prior perfection, automatically.

Joel Lidstrom is Chief Technical Officer of RadMapps.com, and designer of the RadMapps software.