Clinical Validation of Machine Learning Triage of Chest Radiographs

NCT05224479 | Withdrawn

• 1 other identifier: 47832
• Study Type: interventional
• Target: N/A
• Locations: 1 country
• Sites: 1

Brief Summary

Artificial intelligence and machine learning have the potential to transform the practice of radiology, but real-world application of machine learning algorithms in clinical settings has been limited. An area in which machine learning could be applied to radiology is through the prioritization of unread studies in a radiologist's worklist. This project proposes a framework for integration and clinical validation of a machine learning algorithm that can accurately distinguish between normal and abnormal chest radiographs. Machine learning triage will be compared with traditional methods of study triage in a prospective controlled clinical trial. The investigators hypothesize that machine learning classification and prioritization of studies will result in quicker interpretation of abnormal studies. This has the potential to reduce time to initiation of appropriate clinical management in patients with critical findings. This project aims to provide a thoughtful and reproducible framework for bringing machine learning into clinical practice, potentially benefiting other areas of radiology and medicine more broadly.

Conditions

Chest--Diseases

Outcome Measures

Primary Outcomes (1)

• Turnaround time

  Time from completion of radiograph to time that radiologist issues an assessment via preliminary or final report

  up to 1 hour

Study Arms (3)

Traditional workflow triage

ACTIVE COMPARATOR

Radiologists follow standard triage of chest radiographs.

Other: Traditional workflow triage; Other: Machine learning workflow triage; Other: Random workflow triage

Machine learning workflow triage

ACTIVE COMPARATOR

Radiologists follow machine learning triage of chest radiographs.

Other: Traditional workflow triage; Other: Machine learning workflow triage; Other: Random workflow triage

Random workflow triage

SHAM COMPARATOR

Radiologists follow randomly ordered triage of chest radiographs.

Other: Traditional workflow triage; Other: Machine learning workflow triage; Other: Random workflow triage

Interventions

Traditional workflow triage OTHER

Workflow triage is based on order location, STAT designation, and first-in-first-out status.

Machine learning workflow triage; Random workflow triage; Traditional workflow triage

Machine learning workflow triage OTHER

Workflow triage is based on the machine learning model's confidence of abnormality.

Machine learning workflow triage; Random workflow triage; Traditional workflow triage

Random workflow triage OTHER

Workflow triage is based on random order.

Machine learning workflow triage; Random workflow triage; Traditional workflow triage

Eligibility Criteria

• Age: 18 Years+
• Sex: all
• Healthy Volunteers: Yes
• Age Groups: Adult (18-64), Older Adult (65+)

You may qualify if:

• Radiologist at Stanford Hospital and Clinics

You may not qualify if:

• None

Sponsors & Collaborators

• Stanford University: lead
• Society of Thoracic Radiology: collaborator

Study Sites (1)

Stanford University

Stanford, California, 94305, United States

Location

Study Officials

• Emily Tsai, MD

  Stanford University

  PRINCIPAL INVESTIGATOR

Study Design

• Study Type: interventional
• Phase: not applicable
• Allocation: RANDOMIZED
• Masking: SINGLE
• Who Masked: PARTICIPANT
• Masking Details: Radiologists will be blinded when using machine learning and random triage methods.
• Purpose: DIAGNOSTIC
• Intervention Model: CROSSOVER
• Sponsor Type: OTHER
• Responsible Party: PRINCIPAL INVESTIGATOR
• PI Title: Clinical Assistant Professor

Model Details: Radiologists will triage chest radiographs using traditional, machine learning, and random methods.

Study Record Dates

• First Submitted: January 22, 2022
• First Posted: February 4, 2022
• Study Start: August 1, 2022
• Primary Completion: November 1, 2022
• Study Completion: November 1, 2022
• Last Updated: November 1, 2022

Record last verified: 2022-10

Data Sharing

• IPD Sharing: Will not share

Locations

US (1)