Objectives To develop an MRI–based radiomics model for ovarian masses categorization and to compare the model performance to Ovarian-Adnexal Reporting and Data System (O-RADS) and radiologists’ assessments. Materials and methods This retrospective multicenter study included 497 patients (249 benign, 248 malignant) allocated to training, internal, and external validation sets (293/124/80 masses, respectively). Radiomics features were extracted from preoperative MRI. Features were selected using minimum redundancy, maximum relevance, and the least absolute shrinkage and selection operator algorithm. Diagnostic performance of the radiomics model, O-RADS, and independent assessments by junior and senior radiologists was evaluated via the area under the receiver operating characteristic curve (AUC) and compared using DeLong’s test. Results In external validation, the radiomics model (AUC = 0.939) outperformed O-RADS (AUC = 0.862; p = 0.047) and the junior radiologist (AUC = 0.802; p = 0.003) and was similar to the senior radiologist (AUC = 0.886; p = 0.231). Subgroup analysis of O-RADS score 4 showed the model (AUC = 0.879) outperformed both radiologists (junior: p = 0.001; senior: p = 0.005). For solid, cystic–solids, and cystic masses, the AUCs of the model were 0.921, 0.975, and 0.848, respectively. Conclusions The performance of the radiomics model to categorize ovarian masses was superior to O-RADS and junior radiologists and similar to senior radiologists. As a complementary tool to O-RADS, it allows for refined risk stratification for ovarian masses with an O-RADS score of 4 and different morphological characteristics, providing clinicians with quantitative decision support to improve preoperative diagnosis and guide treatment planning. Critical relevance statement Radiomics model provides improved risk stratification and supports precise clinical decision-making for ovarian masses with an O-RADS score of 4 and solid, cystic-solid ovarian masses, thereby improving the management of patients with ovarian masses. Key Points <UnorderedList Mark="Bullet"> <ItemContent> MRI–based radiomics allows for the characterization of ovarian masses with high accuracy. </ItemContent> <ItemContent> Radiomics helps differentiate between benign and malignant ovarian masses with an O-RADS score of 4. </ItemContent> <ItemContent> For solid, cystic–solid, and cystic masses, the radiomics model exhibited higher or similar performance to that of the O-RADS and radiologists. </ItemContent> </UnorderedList> Graphical Abstract

The value of a radiomics model in predicting ovarian malignancy: a retrospective multi-center comparison with O-RADS and radiologists

期刊论文
出版模式：

开放获取
发表日期：

2025年7月31日

查看全文

Junjie Jin,
Xijia Deng,
Ling Long,
Meiling Liu,
Meimei Cao,
Hao Gong,
Huan Liu,
Xiaosong Lan,
Lili Liu,
Jiuquan Zhang

摘要

Objectives

To develop an MRI–based radiomics model for ovarian masses categorization and to compare the model performance to Ovarian-Adnexal Reporting and Data System (O-RADS) and radiologists’ assessments.

Materials and methods

This retrospective multicenter study included 497 patients (249 benign, 248 malignant) allocated to training, internal, and external validation sets (293/124/80 masses, respectively). Radiomics features were extracted from preoperative MRI. Features were selected using minimum redundancy, maximum relevance, and the least absolute shrinkage and selection operator algorithm. Diagnostic performance of the radiomics model, O-RADS, and independent assessments by junior and senior radiologists was evaluated via the area under the receiver operating characteristic curve (AUC) and compared using DeLong’s test.

Results

In external validation, the radiomics model (AUC = 0.939) outperformed O-RADS (AUC = 0.862; p = 0.047) and the junior radiologist (AUC = 0.802; p = 0.003) and was similar to the senior radiologist (AUC = 0.886; p = 0.231). Subgroup analysis of O-RADS score 4 showed the model (AUC = 0.879) outperformed both radiologists (junior: p = 0.001; senior: p = 0.005). For solid, cystic–solids, and cystic masses, the AUCs of the model were 0.921, 0.975, and 0.848, respectively.

Conclusions

The performance of the radiomics model to categorize ovarian masses was superior to O-RADS and junior radiologists and similar to senior radiologists. As a complementary tool to O-RADS, it allows for refined risk stratification for ovarian masses with an O-RADS score of 4 and different morphological characteristics, providing clinicians with quantitative decision support to improve preoperative diagnosis and guide treatment planning.

Critical relevance statement

Radiomics model provides improved risk stratification and supports precise clinical decision-making for ovarian masses with an O-RADS score of 4 and solid, cystic-solid ovarian masses, thereby improving the management of patients with ovarian masses.

Key Points

MRI–based radiomics allows for the characterization of ovarian masses with high accuracy.

Radiomics helps differentiate between benign and malignant ovarian masses with an O-RADS score of 4.

For solid, cystic–solid, and cystic masses, the radiomics model exhibited higher or similar performance to that of the O-RADS and radiologists.

Graphical Abstract