Radiology embraces AI for speed, but hesitates on diagnosis

Voice recognition tools rank as the most widely adopted technology.

Artificial intelligence (AI) adoption in radiology is uneven, with high-impact productivity tools seeing widespread use whilst clinically focused applications trail behind, according to a L.E.K. Consulting report. 

Scheduling and rostering software has the broadest adoption amongst radiologist activities, improving scanner utilisation and resource allocation.

Vendor-native features embedded directly on scanners, covering image acquisition and reconstruction, are also routinely in use, delivering faster scans and reduced radiation exposure.

Voice recognition tools rank as the most widely adopted technology across the entire imaging chain.

One provider CEO told L.E.K. that AI-assisted reporting tools deliver a 30% faster report turnaround for radiologists.

Large language model and report-assist pilots are scaling as a next step, with early users testing generative AI tools to draft near-complete reports for chest X-rays and CT scans, though radiologists still validate and sign off on results.

However, intelligent triage and worklist systems, including Aidoc, Annalise CXR, Arterys, md.ai, Viz.ai, and Rapid AI, are widely used across large networks and teleradiology operations to enable faster urgent diagnosis.

But single-finding computer-aided detection tools, used to pre-analyse scans and highlight suspicious regions, see only mixed day-to-day use, with radiologists required to validate results.

Many radiologists also perceive faster interpretation as being at least partially offset by oversensitivity and false positives.

Communication and admin tools show the widest adoption gap. Closed-loop alert systems are in heavy use in emergency settings, such as stroke pathways, to send alerts directly to neurologists and ER teams.

Outside emergencies, adoption remains uneven, with auto-fill communication tools like Rad AI Continuity generating suggestions based on historical patterns but not yet standard practice.