Bring Your Own AI Builds Clinician Trust and Buy-In

The days when radiologists may have worried that their jobs would become obsolete from the deployment of artificial intelligence (AI) are quickly waning. The potential benefits of AI are just too numerous to ignore—errors of interpretation prevented, diagnostic accuracy improved, stress and burnout reduced—according to a recent opinion piece in the American Journal of Roentgenology. The authors caution, however, that “it is critical that radiologists be intimately involved in guiding the rapid yet careful application of AI for image interpretation...”

And hospitals are attempting to do just that through the careful adoption of AI solutions into their radiology departments. Some hospital systems are building their own AI algorithms, while others are choosing from an ever-expanding library of algorithms from commercial vendors – and others, it is a combination of both. What is clear is a great algorithm is only part of the overall challenge of AI adoption. The promise of AI will only be realized if radiologists trust AI algorithms to be highly accurate and clinically meaningful – and if AI fits seamlessly into their workflow and involves no extra steps to use it.

TeraRecon can help organizations on both fronts. First, by helping data scientists, physicians, and researchers build better algorithms faster with Data Extractor and integrate their algorithms or any purchased standalone algorithm into the radiology department’s existing visualization platform and PACS, with Data Adapter. Both products operate with Intuition 4.5 software. The goal: eliminating the time, frustration, and cost of creating and incorporating AI into radiology workflows and to ensure that physicians have a satisfactory experience using AI.

Building a High-Impact Algorithm

One of the biggest challenges data scientists face in building, training, and validating an AI algorithm is having access to sufficiently large, curated, and representative data. Training an algorithm to identify a lung nodule on a CT scan, for example, may require inputs from 10,000 labeled and annotated lung studies—a costly and impossibly time-consuming process for a team of radiologists to perform manually. But hospitals already have a huge database of images from radiologists who label and annotate CT studies as they interpret them during the ordinary course of routine workflows. This rich source of data, however, isn’t typically in a form that can be readily used to train data.

Data Extractor solves this by easily extracting the valuable segmentation and labeling data that can be used to train the algorithm. Data Extractor can quickly convert archived Intuition data into volumetric, high-fidelity well-annotated data sets, which gives clinicians the confidence that the resulting algorithms will produce highly accurate and valuable insights to assist them in interpreting images. In contrast, clinicians will have little use for algorithms built with limited or poor-quality data.

Making AI Easy to Use

Clinicians will only embrace AI if using the algorithms enable them to be more productive, efficient, and don’t interfere with their normal workflow. Radiologists want to receive AI insights from algorithms that run on the PACS and Advanced Visualization systems, the imaging platform they use continually throughout their workday. If an algorithm can’t be integrated into their normal interpretation workflow, forcing radiologists to log on to a separate imaging platform to use it will only create frustration and add to the problem of physician burnout. Data Adaptor pulls AI inputs generated from any algorithm into Intuition so clinicians can easily get the diagnostic benefits of AI from inside the system they use daily.

Data Adaptor also allows hospitals to scale their AI strategy onto one AI platform called Eureka. While hospital staff may now only be using a handful of AI algorithms, in five years, they may be using 100 algorithms in multiple departments. Deploying scores of algorithms across the entire enterprise, one by one, will involve considerable time, effort, and expense. Data Adaptor allows hospitals to easily build and adopt a library of algorithms onto a platform that is accessible to clinicians throughout the hospital.

Why Do I Need AI?

The longstanding shortage of radiologists has contributed to imaging backlogs at most hospitals, with radiologists facing the impossible task of interpreting one image every 3 to 4 seconds in an 8-hour workday. AI can improve radiologists’ efficiency by identifying suspicious or positive cases for early review. It can lead to faster turnaround and decreased workloads, reducing stress, fatigue, and burnout. AI can improve patient care; with the help of AI, a radiologist can diagnose a brain bleed more quickly, which means the stroke team gets activated faster, and the patient’s treatment is accelerated. The value of AI can be realized in better outcomes for patients and a better quality of work-life for radiologists.