perkinelmer-728×90-20191127-1227
Dr Vasanth Venugopal
Imaging Lead, CARING,
Mahajan Imaging

Implementing AI In Clinical Workflow – How Do We Do It?

These are times when every medical practitioner regardless of their practice setting, field of specialization, or academic interests are bombarded with news articles, lectures, conferences, and textbooks chapters, talking about technical superiority of the new artificial intelligence technologies (especially deep learning) and that how AI will transform the way they practice medicine. But the average practitioner who is hit by the hype for a couple of years now is still missing the impact on his practice ground. So, it is natural for them to ask, How do I get it? In this article, I will explain how we implemented AI in our clinics and how interested radiologists can implement AI solutions at their clinics without affecting their existing workflow.

Live deployments for improving workflow efficiency and accuracy. As you are reading this, chest X-rays from three of our largest diagnostic clinics in Delhi are being analyzed by AI algorithms running on a platform called CARPL, built by our developers. This is a validation, deployment, and analytics platform with rich tools that can enable the radiologist to get on the AI bandwagon on the fly. The platform connected to our PACS automatically queries relevant examinations and processes it through the relevant algorithm. The results are not pushed to the radiology workstations by design as we are still validating them for their accuracy. But instead, these results are compared to the reports made by the radiologist.

The validation setup is used to analyze and quantify the mismatches. It helps us in two ways – first to pick up any misses by the radiologist before the miss can impact the clinical management of the patient, and second to evaluate the performance of an algorithm in near-real time without the radiologist needing to take time out of their clinical work.

Quality-assurance tool. We have also been using AI algorithms as a quality-assurance (QA) tool, where we do a regular retrospective analysis of chest X-rays, including those from our clinics, where we have not yet gone live with our deployments. This helps us to track our errors and implement appropriate corrective measures, including feedback to radiologists and the technologists. We envision this tool to be a standard benchmark to compare performance across our centers and establish minimum working standards for submission to accreditation agencies like NABH.

On-premises AI validation. Given the rapid influx of AI solutions that are being marketed today, it becomes necessary to differentiate the performers from non-performers. This platform also has a data-miner tool that we use to identify and extract relevant cases for testing any new solution. For example, if we need to identify X-rays done for bone-age assessment on a digital radiography machine, we use this search tool on the platform to identify the cases.

The cases are then processed through the AI algorithm that is housed in the platform. These results are then compared to the ground truth through validation analytics tool, and the statistical results are presented to the developer and the validator. This enables us to acquire a broad overview of any algorithm without investing much of our resources.

We are probably one of the few centers that have implemented the high-performance AI tools without having to change our routine workflow. The good news is that we can now bring this platform and its tools to any interested clinic, hospital, or radiologist. If you are interested in deploying AI algorithms in your practice, you know whom to contact.

Share this:
Stay Updated on Medical Equipment and Devices industry.
Receive our Daily Newsletter.