50th Annual OB&GYN Seminar in Orlando, FL and Virtual - Early Bird Rates Now Available! Register Now

TheScan_header_02

Ultrasound Teams Need a Report-Handoff Test Before Using AI

Posted on November 10, 2026

Sonographers, interpreting clinicians, and ultrasound practice managers face a deceptively simple question when they introduce generative artificial intelligence (AI): Who catches an omitted finding before a draft becomes part of patient care? The risk is easy to miss when the generated text reads smoothly and follows the expected format. An accurate report depends on the images, examination context, and clinical judgment behind every sentence.

The AIUM’s April 2026 supervision and interpretation Official Statement places final diagnostic responsibility with the supervising licensed medical provider. That remains the starting point for any workflow that uses AI to summarize examination notes, organize findings, or help prepare communication.

Test the Report Handoff, Not Just the Prompt

Consider three points where AI may appear useful. In a thyroid examination, an assistant might organize measurements and preliminary findings in a narrative. The reviewer must compare every relevant measurement, side, and anatomic descriptor against the actual image record. A polished paragraph is not evidence that the right nodule was described.

In a focused point-of-care examination, an assistant might draft a structured note from dictated observations. The clinician should confirm the clinical question, the limitations of the examination, and what remains unanswered. Language that sounds definitive must not turn a limited examination into a comprehensive one.

In a vascular laboratory, AI may help summarize technical worksheets or draft a message for the ordering team. The reviewing clinician must verify side, segment, measurements, relevant comparison studies, and communication requirements before the message leaves the department.

Build a Four-Field Review Record

A practical starting point is a brief record for each AI-assisted handoff. Document the source material used, the specific task assigned to AI, the professional who checked the output, and any discrepancy corrected before release. Managers can use recurring discrepancies to improve training or adjust which tasks qualify for AI assistance.

That record should remain proportionate to the work. Reformatting an educational handout creates different concerns from drafting a clinical impression. An ultrasound department should define where AI is permitted, when data may be entered, and which outputs require line-by-line source verification.

The AIUM’s practice parameters address examination records, including images and final reports. They provide an existing professional framework for identifying what a generated draft cannot replace: the underlying examination evidence and the responsible provider’s interpretation.

Make Error Reporting Part of Training

Teams often begin with technology demonstrations that reward producing a complete-looking report quickly. A better exercise gives staff an AI-generated draft containing a missing measurement, an incorrect laterality, or a conclusion unsupported by the captured images. Participants must identify the error, explain its significance, and show where the workflow should stop it.

Invite skeptical staff to develop these tests. Their familiarity with common documentation problems can reveal failure modes that enthusiastic early adopters overlook. Encourage staff to report AI mistakes without treating every discovery as personal failure. The NIST AI Risk Management Framework offers a useful general structure for assigning responsibility, identifying risks, measuring outcomes, and managing them.

Measure the Check, Not Just the Time Saved

During a small pilot, track both time spent and verification outcomes. How often did a reviewer correct an omission, wrong measurement, ambiguous limitation, or unsupported conclusion? Did the process improve communication and consistency without increasing the burden on interpreting clinicians? A workflow that produces faster drafts but requires extensive repairs has not earned broader use.

Start with a bounded task, a defined professional reviewer, and an agreed stopping rule. Once the team can demonstrate reliable source-to-report checks, managers can decide whether a larger pilot is warranted. Patients depend on the accuracy of the final communication, regardless of who or what helped prepare the first draft.

 

Adapted from: The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026) https://disasteravoidanceexperts.com/aibook.

 

Dr. Gleb Tsipursky, a behavioral scientist called the “Office Whisperer” by The New York Times, helps tech-forward leaders stop overpaying for AI while boosting engagement and innovation. He serves as the CEO of the AI consultancy Disaster Avoidance Experts, and wrote eight books, including The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). 

Photo credit: Courtesy of Dr. Gleb Tsipursky.

Authors and Contributors