Promises and Perils of AI – Part 29: Machine Vision: Seeing the Future with AI Imaging

This week I explore the question: How is AI revolutionizing diagnostic imaging and interpretation?

Medical imaging—across radiology, pathology, dermatology, and ophthalmology—is central to diagnosing a vast array of conditions. Yet traditional interpretation is time-intensive, subjective, and dependent on highly trained specialists. Enter AI-powered machine vision, where deep learning models trained on millions of images can now identify, classify, and quantify disease features with unprecedented accuracy and speed.

These models are transforming static images into dynamic diagnostic tools. AI doesn’t just detect abnormalities—it tracks disease progression, measures treatment response, and flags urgent findings in real time. Algorithms can also be trained to identify conditions that are frequently missed or misdiagnosed, such as early-stage lung cancer on chest X-rays or diabetic retinopathy in fundus photographs.

Importantly, regulatory approval of these tools is growing. The FDA has cleared over 500 AI/ML-enabled medical devices as of 2024, many of them focused on imaging. As these tools become embedded into diagnostic workflows, they are reshaping how we deliver and pay for image-based care.

For Providers: AI-assisted imaging enhances productivity and quality across clinical specialties. Radiologists are using AI to triage studies, prioritize critical findings (like brain hemorrhages or pulmonary emboli), and reduce diagnostic variability across readers. In pathology, AI helps quantify tumor burden or grade biopsies with objective metrics.

Companies like Aidoc provide real-time AI triage solutions that integrate directly into radiologist workflows and have shown to reduce turnaround times by over 30%. Arterys (now part of Tempus) delivers cloud-based AI for cardiac and lung imaging, enabling seamless, collaborative diagnosis.

In underserved or rural areas, AI systems act as virtual specialists—offering near-instant interpretation for X-rays or retinal scans where human expertise may be limited. EyeArt by Eyenuk, for instance, is an FDA-cleared solution for autonomous detection of diabetic retinopathy in primary care settings.

Other leading players for providers:

• PathAI – AI-powered pathology for more accurate cancer diagnostics.

• RadNet’s DeepHealth – Specializes in AI for breast imaging and mammography.

For Payers: Payers are increasingly turning to AI imaging analytics to ensure medical necessity, speed up claims adjudication, and reduce unnecessary imaging utilization. By analyzing the content of images—not just documentation—AI helps validate clinical claims and support pre-authorization decisions with data-driven transparency.

AI can also drive significant cost savings through earlier detection of disease, reducing downstream spending on advanced-stage interventions or avoidable complications. For example, detecting lung nodules or aortic aneurysms earlier can shift patients into lower-cost, higher-quality treatment pathways.

Companies like Cleerly use AI to detect coronary artery disease in CT angiography and provide quantitative plaque burden analysis, which payers can use for risk stratification and preventive care planning. Similarly, Qure.ai works with insurers globally to provide image-based risk insights, especially in tuberculosis and stroke care.

Other notable companies working with payers:

• Nanox.AI – Focuses on population screening and chronic disease detection from routine imaging.

• Perspectum – Delivers imaging biomarkers for chronic liver disease and metabolic risk management.

Next Steps/Recommendations:

1. Integrate AI tools directly into PACS/RIS and radiology workflows for seamless use by clinicians.

2. Ensure AI explainability and transparency in decision-making to foster trust and reduce liability.

3. Promote payer-provider collaborations to align AI interpretation with value-based care models and payment innovation.

Philosophical Question to Ponder in Closing: If AI can outperform radiologists in some tasks, how should we redefine the role of human expertise in diagnostics—judgment, empathy, oversight, or something entirely new?

Disclaimer: The views and opinions expressed in this post are solely my own and do not necessarily reflect the views or positions of my employer or any affiliated organizations.