Artificial intelligence-based sequential ultrasound-MRI strategy for ovarian masses：dual evaluation of diagnostic accuracy and healthcare costs
10.3760/cma.j.cn131148-20250410-00202
- VernacularTitle:基于人工智能的卵巢肿块超声-磁共振序贯诊断策略：诊断准确性与医疗成本的双重评价
- Author:
Jingjing YU
1
;
Ruixia DAI
;
Xiaomin LIU
;
Peijun HU
;
Xiaochen WANG
;
Sihui HU
;
Shanshan ZHANG
;
Wenqian WANG
;
Yu TIAN
;
Jiale QIN
Author Information
1. 浙江大学医学院附属妇产科医院超声科，杭州 310006
- Publication Type:Journal Article
- Keywords:
Ultrasonography;
Ovarian cancer;
Artificial intelligence;
Magnetic resonance imaging;
Sequential diagnosis
- From:
Chinese Journal of Ultrasonography
2025;34(9):759-765
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop an artificial intelligence（AI）-based sequential ultrasound-magnetic resonance imaging（US-MRI）diagnostic strategy to optimize the imaging workflow for ovarian masses.Methods:A total of 1 120 patients with pathologically confirmed ovarian masses who underwent both preoperative pelvic ultrasound and MRI between January 2021 and December 2023 at Women's Hospital，Zhejiang University School of Medicine were retrospectively included. Patients were randomly divided into the training（ n=672）and internal test set（ n=448）at a ratio of 6∶4. An external test set（ n=128）was established at the Forth Affiliated Hospital of School of Medicine. Deep learning was used for automated segmentation of MRI lesions，followed by radiomic feature extraction and machine learning classification to construct both a US-MRI multimodal model and sequential US-MRI strategy. Diagnostic performance and potential healthcare cost-saving effects were evaluated across strategies. Results:In the internal test set（ n=448），the AI-based sequential US-MRI strategy achieved a F1 score of 0.863 and a diagnostic accuracy of 82.14%，with no significant difference compared to the US-MRI multi-modal model（ P>0.05）. The sequential strategy identified 82 cases（18.30%，82/448）of patients as low-risk true negatives during initial ultrasound screening，suggesting a potential to reduce the need for MRI examinations in future clinical practice. In the external test set（ n=128），the strategy achieved an F1 score of 0.800 and a confirmed diagnosis rate of 85.94%，with a theoretical reduction of 26.56%（34 cases）in MRI utilization while maintaining a diagnostic accuracy rate higher than that of the multi-modal model（82.18%）. Conclusions:The AI-based US-MRI sequential diagnostic strategy demonstrates favorable diagnostic accuracy while offering the potential to optimize MRI utilization. This approach may enhance the efficiency of imaging resource allocation and reduce healthcare burden in the management of ovarian masses.
