Three Steps to Approach Adoption of AI to Enhance Medical Imaging Services

  • October 04, 2022
doctor reviewing scans on screen

The opportunity to adopt artificial intelligence (AI) in medical imaging was once a future-facing goal and is now a requirement for healthcare providers. With so much rapid change in the industry, any organization that neglects to incorporate AI technology runs the risk of being left behind in value-based clinical care and cost optimization.

The expectation of AI in healthcare is precision and expertise. Adopting AI for medical imaging should exceed or minimally match the everyday experience of patients and consumers of imaging services. AI promises healthcare providers cost-savings and workflow optimization. However, it’s not as simple as integrating any AI technology into your organization.

Regulatory approval of medical imaging AI does not mean that AI will meet expectations or produce results consistent with existing practices. Experience is created by equipment, protocols, patients, and the interpretations across the practices of diagnosticians. The collective staff experience reflects their training and knowledge of interpreting cases for their patients and clientele.

Any adoption strategy should align practice challenges and department priorities for revenue growth and investment return. So how do you adopt AI to maximize outcomes while improving your bottom line? Here are three priorities when implementing an AI adoption strategy for medical imaging:

1. Identify where AI will have the most impact

The first step is to formulate a vision, take an in-house inventory of available AI solutions and identify your priorities. The priority of AI adoption may be addressing current AI inventory or an investment strategy around new AI. In all cases, there is a need to realize some level of ROI as adopting AI has costs for platforms, algorithms, and process implementation.

Recent surveys set high expectations for AI adoption, with the statistics that 68% to 82% of health system execs in the U.S. plan deeper AI investments to meet strategic goals. Rising staff costs and reduced reimbursement are eating away at investment strategies, so the ROI timeframe is shorter. This requires focusing limited resources to plan, implement and monitor to ensure optimal returns.

Planning for AI considers which tools may best suit your patient population and professional service lines and how the value of each algorithm and its outputs improve the overall care delivery process. The strategic plan must account for how AI impacts the ‘people’ component and how they can add value through AI adoption.

2. Aim for transparency and compliance with current practice

The key to accepting and adopting AI begins with understanding how AI makes decisions, how it is trained, and how it describes findings. Transparency of the deliverable allows the diagnostician to feel comfortable explaining the outcome of an AI model and can communicate the results effectively.

Many vendors have imaging AI tools to explain the AI outcome using different techniques, including exposure of decision trees, annotated anomalies, and heat maps. However, images are not binary, so these graphic tools can lead to increased time spent understanding the outcome. Identifying differences between AI and ground truth outcomes, and estimating their implications, requires clinical validation of ‘expertise,’ which requires support and staffing not always available in-house.

One measure is the sensitivity and specificity — the true positive and true negative rates matching the outcomes of the existing human system. Another tool uses the positive predictive value (PPV) to describe where the positive prediction is genuinely positive. A lower PPV describes the risk of a positive finding being false.

While not all methods are described here, the baseline is a representative cohort of your institution or practice’s patient population that has been curated to remove the API. The second asset to evaluation is navigating how the different algorithm (AI) create their prediction/recommendation.

3. Set expectations for ROI

Conceptually how AI improves the value of care is one definition of return on investment (ROI). Without equivocation, there is a cost to adopting AI that may or may not be addressed through reimbursement in the U.S. versus other national health program markets. Understanding the value within an ROI begins with the impact on diagnostic reporting leading to costs that can be avoided — such as false positives, early detection/health screening, the impact of incidental findings, and care compliance.

Given the ‘expertise’ of each AI tool in specific environments, early indications show significant opportunities to create savings in physician workflow by running multiple AI systems in tandem using a platform approach to AI. Any reimbursement per use as a justification for ROI is far from certain, and relying on outcomes is longer-term and hard to measure. With the pressure on funding, other areas of savings, once incremental benefits, become a focus of the model. Another example is how AI in medical imaging impacts the automation of tasks for clinicians and acuity in detection leading to the consistency of diagnosis and, thus, patient outcomes. AI automation can also support new clinical outreach efforts and be core to creating centers of excellence.

Creating an adoption strategy

One powerful service that helps facilitate the three steps above is NTT DATA’s Advocate AI offering. Our tools and team have proven experience in various healthcare and medical imaging use cases. Our practical methodologies help clients match AI to their unique needs and practices and identify opportunities in their environment for ROI.

Using internally developed processes, Advocate AI creates an analysis of your patients and practice that enables insights into where AI could be applied. In later stages, the service provides the basis for monitoring the performance of new software releases for the AI tools and baseline existing tools against new or emerging AI offerings.

Talk to us about our Advocate AI services to find out more about how you can reduce evaluation and investment request cycles with a comprehensive blueprint to accelerate your adoption of AI.

Subscribe to our blog

Mitchell Goldburgh

Mitchell Goldburgh is a Global Solutions Leader for NTT DATA Services' Enterprise Imaging and Analytics practice. A 35-year veteran of healthcare imaging, he has held a variety of roles in the provider segment, as well as healthcare business development positions for public and startup technology companies. Mitchell has served as co-chair of numerous healthcare standards committees, and authored chapters in academic journals and books on digital imaging. In his current role, Mitchell provides management around NTT DATA’s participation in the evolution and adoption of digital imaging in healthcare, and has driven NTT DATA resources around the integration of analytics for imaging.

Related Blog Posts