Incorporating computer vision into a clinical decision support system with systems analysis


AI is often shrouded in mystery due to the hype around it. The reality is that AI is neither magical nor unobtainable. AI has the potential to revolutionize health care but these systems are challenging to implement. The key to successfully integrating any new technology, including AI, into an organization is based on good old-fashioned systems analysis, which requires understanding both your current system and your desired system. This approach can be especially useful for AI projects because the integration of AI into the system is clearly documented and can be used in discussions with the stakeholders involved in the project.

I recently had an opportunity to peek into a systems analysis project for the design of a clinical decision support system (CDSS) which integrates computer vision to help radiologists monitor the progression of tumors. I spoke with Antonie Berkel and Dr. Joost Huiskens from our SAS Netherlands office about this project.

To begin this discussion, what is a clinical decision support system (CDSS) and why does it matter to clinicians?

A clinical decision support system is a tool that can be used to augment a clinician’s decision making by integrating large volumes of information and from that, provide recommendations for patient care. The advantage of a CDSS is that it can help deliver more personalized medicine to patients by focusing on individual needs and preferences.

What is the clinical case study used in this systems analysis?

Amsterdam University Medical Center (AUMC) provides patient care, conducts scientific research, and is also a teaching hospital. They are monitoring the progression of tumors in the liver to see if a tumor has shrunk or if its appearance has changed. The results of this evaluation help determine the next steps for a patient’s plan of care, such as surgery or the application of a different chemotherapy regimen (drugs, dosages, timing) or other alternatives.

What challenges do you see when implementing these systems?

The first step in the systems analysis was to understand the challenges that were commonly encountered when designing a CDSS. While there are many challenges encountered during the design of a CDSS, the key challenges are integration with clinician workflow and integration with an existing hospital systems infrastructure. This means that AI should integrate with both the existing systems for patient care and the clinician’s workflow.

In your example, what is the current workflow that clinicians follow?

For this case study, a general clinical workflow starts with the treating physician ordering a CT scan. These images are acquired by a machine as set up by a CT technician. Then a radiologist reviews the images and provides an interpretation that is included in a report. The report is included in a discussion of the patient’s case by a multidisciplinary medical team to formulate the patient’s plan of care.

The problem is that the evaluations of the CT scans are very workload intensive because

  • Evaluating tumors is a time-consuming process for radiologists.
  • And for each CT scan, typically only the two largest tumors are measured – possibly missing information hidden in any remaining tumors.
  • Furthermore, the manual assessment is prone to subjectivity, which may result in different evaluations amongst a team of radiologists. Being able to make this assessment more objective will also improve patient care.

What do you think is the most critical output from your systems analysis?

The critical output from this system analysis is the creation of an enterprise reference architecture that illustrates a layered overview of the integration between business processes, software applications, and medical infrastructure. Understanding these three layers is the foundation for developing the behaviors of the desired CDSS which in this case incorporates AI capabilities such as computer vision.

What is the role of computer vision in your desired system?

AUMC is using computer vision from SAS to monitor the progression of tumors in the liver starting with DICOM images from CT scans. DICOM stands for Digital Imaging and Communications in Medicine. The project started by training a deep learning model with data from 52 cancer patients. Every pixel of 1,380 metastases was analyzed and segmented. This taught the system how to instantly identify tumor characteristics and facilitate a chemo response assessment.

Are there any standards in radiology that you needed to take into consideration?

There are several standards that need to be considered in this implementation of a CDSS. The three key standards include DICOM, HL7, and IHE. DICOM is the standard for the communication and management of medical imaging related data. HL7 is the set of international standards for the transfer of clinical and administrative data between software applications used by various healthcare providers. IHE is an initiative by healthcare professionals and industry to improve the way computer systems in healthcare share information and support workflows

What are the key changes that you made to the existing workflow to create your desired system?

Clinicians have the option in their healthcare information system (HIS) to request an automated assessment of a tumor to augment their own assessment. This is considered as a separate study from the one performed by the radiologist. This preserves the existing clinical workflow while being integrated within the existing HIS. Having a separate study augments the radiologist and helps address the workload intensive nature of the assessment to ultimately improve patient care.

How do see CDSS’ that utilize AI affecting patient care?

Computer vision, incorporated into a CDSS, can augment radiologists’ decision making and make the image interpretation process more cost-effective, faster, and more accurate. The ultimate goal is to achieve a better patient outcome facilitated by the use of computer vision.

Any advice to those who want to employ a systems engineering approach to the integration of AI into their existing workflows?

AI is not something that can be grafted into a CDSS. There needs to be time built into the development cycle of the CDSS to understand how AI fits into both the clinician workflow and the hospital’s infrastructure to increase the chance of the system being implemented. Once the systems analysis has been completed, a reference architecture can be built which incorporates clinical domain knowledge, healthcare industry standards, and AI capabilities inside the clinician workflow to augment the decision making.

Thanks to Antonie and Joost for sharing this very practical and well thought out example of systems analysis.

To learn more, you can visit these resources:


About Author

Susan Kahler

Global Product Marketing Manager for AI

Susan is a Global Product Marketing Manager for AI at SAS. She has her Ph.D. in Human Factors and Ergonomics, having used analytics to quantify and compare mental models of how humans learn complex operations. Throughout her well-rounded career, she has held roles in user centered design, product management, customer insights, consulting and operational risk. Susan recently completed her Master of Science in Analytics, focusing on healthcare analytics. She also holds a patent for a software navigation system to guide users through dynamically changing systems.

Related Posts

1 Comment

Back to Top