Automated Reporting System for Biliopancreatic Endoscopic Ultrasound

This study aims to develop and validate an automated reporting system which will generate structured preliminary reports based on preprocedural clinical data, the endosonographer's speech during EUS, EUS videos, and EUS procedural records. System development will be completed before prospective validation, using a dataset separated from the prospective validation cohort.

The prospective multicenter validation will enroll 420 adults undergoing EUS for suspected or confirmed biliopancreatic diseases. Each patient will contribute only the first eligible EUS examination as the index EUS examination. For each index examination, the automated reporting system will generate a system-generated preliminary report, while the examining endosonographer will independently generate the manual preliminary report without AI assistance. Additionally, speech-only preliminary reports and video-only preliminary reports will be generated to explore the contribution of different data inputs.

Each report will be independently assessed by two EUS experts blinded to the report source. Disagreements between the two assessors will be adjudicated by a third expert. Significant clinical corrections are defined as corrections to errors or omissions that could affect lesion localization, diagnosis, or subsequent clinical management; word and format changes alone do not qualify as significant clinical corrections. Clinical acceptance rate is defined as the proportion of all reports generated for enrolled patients that were accepted without significant clinical corrections. The primary endpoint is the difference in clinical acceptance rates between the system-generated preliminary report and the manual preliminary report of the same patient.

Additionally, a predefined nested endosonographer-reporting substudy will use the same cases to compare completion time and quality between final reports prepared by endosonographers with and without assistance of a system-generated preliminary report. Reporting tasks will be randomly assigned to endosonographers. This substudy will evaluate whether the automated reporting system can improve reporting efficiency while maintaining final report quality.

Adults undergoing EUS for suspected or confirmed biliopancreatic diseases will participate in this study. Each patient provides only the first eligible EUS examination for a prospective validation of the automated reporting system.

Device: Automated Reporting System for Biliopancreatic Endoscopic Ultrasound

The system that integrates preprocedural clinical data, the endosonographer's speech during EUS, EUS videos, and EUS procedural records to generate a system-generated preliminary report. The system-generated preliminary report is generated for research validation after the routine EUS examination and is compared with the endosonographer's independent manual preliminary report. The system does not determine the patient's clinical care.

Time frame: Preliminary reports from the index EUS examination; assessment within 60 days after EUS.

The within-patient paired difference in clinical acceptance rates, calculated as the rate for system-generated preliminary reports minus that for independently written manual preliminary reports. A report is accepted if it requires no significant clinical corrections-corrections to errors or omissions that could affect lesion localization, diagnosis or clinical management. Failed or blank system-generated reports are classified as not accepted. Non-inferiority is established if the lower limit of the two-sided 95% confidence interval is greater than -5 percentage points.

Time frame: Index EUS examination; report assessment within 60 days.

In the prospective paired validation, completeness is the proportion of applicable, reference-confirmed items correctly documented in each system-generated preliminary report. Items cover actual examination extent, organ findings, lesion descriptions, and sampling records.

Time frame: Index EUS examination; report assessment within 60 days.

In the prospective paired validation, the number of significant clinical corrections required per system-generated preliminary report will be assessed. Reductions and additions relative to the independent manual preliminary report will also be reported.

Time frame: During the first automated report-generation task for the index EUS examination, within 30 days after EUS.

In the prospective paired validation, time is measured from submission of all required inputs-including preprocedural clinical data, the endosonographer's speech, EUS videos, and EUS procedure records-to availability of the first system-generated preliminary report. Failed generation will be reported separately.

Time frame: During the routine reporting session immediately after the index EUS examination, up to 60 days

Description: In the prospective paired validation, time is measured from the endosonographer beginning routine report preparation to submission of the independent manual preliminary report, without AI assistance.

Time frame: During the assigned reporting tasks for each index EUS examination, up to 60 days

In the randomized endosonographer-reporting substudy, the within-case difference in total time to complete a final report is calculated as unassisted time minus assisted time. Total time includes review of source materials, report generation waiting time where applicable, writing, and verification. This outcome will be formally tested only if non-inferiority of the primary outcome is established.

Time frame: Final reports from the assigned reporting tasks for each index EUS examination; assessment within 60 days after EUS.

In the randomized endosonographer-reporting substudy, the proportion of final reports accepted without significant clinical corrections will be assessed separately for reports completed with and without access to the system-generated preliminary report. Completeness and clinically important errors and omissions will also be reported.

Development and Prospective Multicenter Validation of an Automated Reporting System for Biliopancreatic Endoscopic Ultrasound