As healthcare continues to evolve, the industry is rife with AI solutions that promise to solve a host of challenges from operational efficiency to clinical diagnoses. With so many potential vendors, how does one determine what solution is right for their organization? Here are 5 questions to ask yourself before investing in enterprise AI solutions:
1. HOW WILL THIS SOLUTION IMPACT MY CLINICIANS’ DAY TO DAY?
Successful application of AI should not be disruptive; rather, it should be complementary and embedded into the current workflows of the systems used daily. Clinicians should be able to enjoy the benefit of AI without having to launch multiple windows, viewers, or widgets.
2. WHAT ARE MY MINIMUM REQUIREMENTS FOR PLATFORM PERFORMANCE?
AI Platforms are not all equal and there are a few minimum requirements that you should consider. For example, in medical imaging, an AI platform should:
- Use standards-based communication and APIs for connecting with PACS, VNA, and AI algorithm endpoints.
- Orchestrate the application of AI algorithms and studies at the procedure, series, and image level.
- Include load balancers to ensure servers can handle the workload.
- Apply AI models and algorithms as needed to the worklist or viewer using standards or API level integrations.
- Provide a dashboard that shows logs, activity, model insights, and performance and productivity analytics or metrics.
- Standardize study data to ensure orchestration, hanging protocols, and routing works as designed.
- Provide insights from historical clinical data instantly, in the same interface and workflow as net new data.
- Enable generation of real-world evidence for population health analysis, research, quality assurance, and education.
3. DOES THIS SOLUTION PROVIDE MECHANISMS FOR APPROVAL AND REJECTION BY THE DICTATING CLINICIAN?
In some cases, harm can be done to patients if a clinician takes an AI finding out of context. Quality control and accuracy of diagnoses are at the forefront of the continuum of care. As such, the chosen AI solution should include ways for clinicians to audit and approve insights.
4. HOW SECURE IS THIS SOLUTION?
Careful consideration should be taken with regard to PHI (Personal Health Information) safeguards. Cloud deployments should be vetted to mitigate risk. Anonymization and encryption should be mandatory.
5. HOW CAN I ENSURE THAT THE AI ALGORITHMS ARE ACCURATE?
AI algorithms should be monitored for accuracy after initial implementation. Quality can sometimes degrade or drift due to differences in an algorithm’s training data versus a healthcare organization’s production data. For example, patient population data from an older age group may show different health characteristics than patient population data from a younger age group. An algorithm trained on data from one patient population may require retraining before being applied to another population.
Regulatory agencies closely monitor medical devices to ensure performance across patient populations.
Additional post-market surveillance by the user can further ensure the desired safety and operating characteristics.