An AI-Assisted Agentic System for Ultrasound Scanning and Diagnosis of Ovarian Lesions

Condition(s) studied

Investigational drug(s) / intervention(s)

non-AI assisted and then AI assisted: Participants first undergo ultrasound scanning and diagnosis by a junior sonographer without AI assistance, followed by ultrasound scanning by the same sonographer with AI assistance.

AI assisted and then non-AI assisted: Participants first undergo ultrasound scanning and diagnosis by a junior sonographer with AI assistance, followed by ultrasound scanning by another junior sonographer without AI assistance. The two junior sonographers are blinded to each other's scanning and assessment results.

Study summary

Investigators developed an interactive agentic system designed to guide newly qualified sonographers in ovarian lesion scanning and improve their scanning quality and diagnostic performance toward expert-level standards. Our agentic system is capable of capturing key features including the max-diameter plane of ovarian lesions from dynamic ultrasound videos, translating these findings into standardized International Ovarian Tumor Analysis (IOTA) descriptors, and providing multi-turn guidance for subsequent scanning, and ultimately generating an AI-assisted diagnostic assessment based on embedded expert knowledge.

In this multicenter study, participants are asked to undergo gynecological ultrasonography performed by sonographers with less than 3 years of experience with or without AI assistance. Our researchers will compare the performance of operators working with AI against that of operators working without AI, as well as against the performance of expert sonographers, to see whether AI assistance enhances the proficiency of less experienced operators and help them approach the scanning quality and diagnostic accuracy of expert sonographers in real-world clinical scenarios.

Eligibility

Primary outcome measure(s)

• Completeness of ultrasound feature acquisition — Up to 7 days from completion of the study
  The completeness of ultrasound feature acquisition will be assessed with reference to national authoritative quality-control standards. After each examination, all stored ultrasound images and videos will be labeled according to their intended purpose. A feature will be considered adequately acquired when at least one stored image or video provides sufficient visual evidence for assessment of that feature. The proportion of required features successfully acquired will be calculated. The proportion of redundant stored images will also be recorded as an additional indicator of acquisition quality.
• Accuracy of interpretation of individual ultrasound features — Up to 7 days from completion of the study
  The accuracy of interpretation of ultrasound features will be assessed by comparing the assessments made by junior sonographers under AI-assisted and unassisted conditions with the findings independently acquired and interpreted by expert sonographers during a separate repeat ultrasound examination. The accuracy rate for each feature will be calculated.
• Agreement between junior and expert sonographers in ultrasound diagnosis — Up to 7 days from completion of the study
  Agreement between junior and expert sonographers will be assessed for the final diagnosis of ovarian lesions. The assessments made by junior sonographers under AI-assisted and unassisted conditions will be compared with the corresponding diagnoses independently obtained by expert sonographers during a separate repeat ultrasound examination, which will then be quantified using appropriate agreement statistics.
• Diagnostic accuracy of ovarian tumors — Within 3 months after the ultrasonography examination
  Diagnostic accuracy will be assessed against histopathological diagnosis as the golden reference standard. This outcome will be evaluated only in participants who undergo surgical treatment and have definitive histopathological findings based on paraffin-embedded surgical specimens.
• Diagnostic performance of ovarian tumors — Within 3 months after the ultrasonography examination
  Area under curve will be assessed against histopathological diagnosis as the golden reference standard. This outcome will be evaluated only in participants who undergo surgical treatment and have definitive histopathological findings based on paraffin-embedded surgical specimens.
• Diagnostic performance of ovarian tumors — Within 3 months after the ultrasonography examination
  Diagnostic sensitivity and specificity will be assessed against histopathological diagnosis as the golden reference standard. This outcome will be evaluated only in participants who undergo surgical treatment and have definitive histopathological findings based on paraffin-embedded surgical specimens.
• Diagnostic performance of ovarian tumors — Within 3 months after the ultrasonography examination
  Diagnostic F1-score will be assessed against histopathological diagnosis as the golden reference standard. This outcome will be evaluated only in participants who undergo surgical treatment and have definitive histopathological findings based on paraffin-embedded surgical specimens.

Trial sites (2)

More Women's Hospital School Of Medicine Zhejiang University trials in China

Other trials for the same condition