All Quiet on the Clinical Front -- Tales from Crossing the Great Divide
Poster No:
2250
Submission Type:
Abstract Submission
Authors:
Rudolph Pienaar1,2, Jennings Zhang1, Sandip Samal1, Gideon Pinto1, Ellen Grant3
Institutions:
1Boston Children's Hospital, Boston, MA, 2Harvard Medical School, Boston, MA, 3Division of Newborn Medicine, Boston Children’s Hospital and Harvard Medical School, Boston, MA
First Author:
Co-Author(s):
Ellen Grant
Division of Newborn Medicine, Boston Children’s Hospital and Harvard Medical School
Boston, MA
Division of Newborn Medicine, Boston Children’s Hospital and Harvard Medical School
Boston, MA
Introduction:
Integrating the results of research workflows into clinical systems can have many beneficial effects. For the most part, however, this final connection is rarely discussed or investigated in the literature and is almost left as an exercise for the reader. While most researchers might code complex computational workflows that provide meaningful scientific and clinical results, these same researchers rarely consider how to bridge the clinical / research divide.
In this work, we summarize some reasons for this status quo, based on our own experiences as well as detailed interactions with clinicians at our Institution. Based on these insights, we importantly also present an open-source solution architecture that does bridge this divide. While aspects of this design are specific to our research back end, the solution itself is easily generalizable. We will discuss several workflows that we are integrating from research into clinical domains.
In this work, we summarize some reasons for this status quo, based on our own experiences as well as detailed interactions with clinicians at our Institution. Based on these insights, we importantly also present an open-source solution architecture that does bridge this divide. While aspects of this design are specific to our research back end, the solution itself is easily generalizable. We will discuss several workflows that we are integrating from research into clinical domains.
Methods:
Most of our insights stem from an internally funded award tasked with providing the results of a research-based AI bone analysis tool to clinicians. This is a non-FDA approved tool that merely aids in clinical decision making by saving clinicians from having to make manual measurements on a bone image. We have conducted numerous interviews with clinicians to better understand their workflow and appetite for value added (AI) tools, similarly we have experience with fellow researchers and their approach to this problem.
Based on these insights, we have determined two possible solutions: (1) a push solution where a clinical system "initiates" the integration; and (2) a pull solution in which a research system polls for data and self-initiates all processing.
In this work we describe (1). The push solution has to be as simple as possible, see Figure 1. Since many clinical systems provide "user-configurable triggers" in their UI, we decided that the simplest possible connection is a single http POST request to a web address. This POST request contains information about the image to be analyzed (such as the SeriesInstanceUID). The return payload contains information on the progress of the analysis.
Based on these insights, we have determined two possible solutions: (1) a push solution where a clinical system "initiates" the integration; and (2) a pull solution in which a research system polls for data and self-initiates all processing.
In this work we describe (1). The push solution has to be as simple as possible, see Figure 1. Since many clinical systems provide "user-configurable triggers" in their UI, we decided that the simplest possible connection is a single http POST request to a web address. This POST request contains information about the image to be analyzed (such as the SeriesInstanceUID). The return payload contains information on the progress of the analysis.
Results:
From the clinical side, we determined there is little to no interest to changing or supplementing existing workflows, even if such a change could have benefits. Simply stated, we noted that clinicians are opposed to even opening a new browser and interacting with a new website. Any integration has to happen without as little friction as possible. From the research side, coding complex infrastructural software is not seen as advancing the research enterprise.
Moreover, there is a prevailing belief that integration between clinical and research is best left to commercial companies. In our experience, this belief is naive and has the practical effect of little to no integration ever happening. We believe that researchers can successfully lead and contribute to building practical solutions -- especially since researchers themselves are often times embedded in clinical hospitals.
Our solution is "fire-and-forget". The clinical system triggers an analysis by simply accessing a web URL. This is a server that based on the trigger event, further interacts with a more complex command and control center that in turn is able to pull data from a clinical system and pass to our research computing system, ChRIS, where a full analysis is managed. Importantly, one of the research nodules is able to itself connect back to the PACS and push resultant images to PACS.
Each time the clinical system sends the same trigger event to the bridge server, it receives a status update (see Figure 2).
Moreover, there is a prevailing belief that integration between clinical and research is best left to commercial companies. In our experience, this belief is naive and has the practical effect of little to no integration ever happening. We believe that researchers can successfully lead and contribute to building practical solutions -- especially since researchers themselves are often times embedded in clinical hospitals.
Our solution is "fire-and-forget". The clinical system triggers an analysis by simply accessing a web URL. This is a server that based on the trigger event, further interacts with a more complex command and control center that in turn is able to pull data from a clinical system and pass to our research computing system, ChRIS, where a full analysis is managed. Importantly, one of the research nodules is able to itself connect back to the PACS and push resultant images to PACS.
Each time the clinical system sends the same trigger event to the bridge server, it receives a status update (see Figure 2).
Conclusions:
In conclusion, we believe this bridge solution is a viable concept for crossing the clinical/research divide and accelerating the ability to have research solutions provide value added benefit to clinical work.
Neuroinformatics and Data Sharing:
Databasing and Data Sharing
Workflows 2
Informatics Other 1
Keywords:
Computing
Data analysis
Workflows
1|2Indicates the priority used for review