Development and validation of an interpretable model integrating multimodal information for improving ovarian cancer diagnosis

Huiling Xiang ^1,2, Yongjie Xiao ³, Fang Li ^4,5

¹Department of Radiology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou, 510060, P. R. China.
²Department of Ultrasound, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou, 510060, P. R. China.
³AI Research Lab, Imsight Technology Co., Ltd., Nanshan, Shenzhen, 518000, China.
⁴Department of Ultrasound, Chongqing University Cancer Hospital, Chongqing, China.
⁵Chongqing Key Laboratory for Intelligent Oncology in Breast Cancer (iCQBC), Chongqing University Cancer Hospital, Chongqing, 400030, China.
⁶Department of Ultrasound, Guangdong Second Provincial General Hospital, No. 466, Xingang Middle Road, Haizhu District, Guangzhou, Guangdong, China.
⁷Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Hong Kong, China.
⁸Zhejiang Lab, Hangzhou, China.
⁹Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, China.
¹⁰Department of Ultrasound, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou, 510060, P. R. China. linxi@sysucc.org.cn.
¹¹Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Hong Kong, China. jhc@cse.ust.hk.
¹²Department of Chemical and Biological Engineering, The Hong Kong University of Science and Technology, Hong Kong, China. jhc@cse.ust.hk.

⁰Department of Radiology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou, 510060, P. R. China.

Nature Communications +

|

March 28, 2024

View abstract on PubMed

Summary

OvcaFinder, an interpretable AI model, enhances ovarian cancer diagnosis by integrating deep learning, radiologist scores, and clinical data. It significantly improves diagnostic accuracy and consistency for radiologists.

Area Of Science

Oncology
Medical Imaging
Artificial Intelligence

Background

Ovarian cancer is a leading cause of gynecological cancer mortality.
Accurate and early diagnosis is critical for improving patient outcomes.
Current diagnostic methods require enhancement for better accuracy and consistency.

Purpose Of The Study

To develop and validate OvcaFinder, an interpretable AI model for ovarian cancer diagnosis.
To assess the performance of OvcaFinder compared to existing clinical and deep learning models.
To evaluate the impact of OvcaFinder on radiologist diagnostic accuracy and inter-reader agreement.

Main Methods

Developed OvcaFinder by combining deep learning predictions from ultrasound images, Ovarian-Adnexal Reporting and Data System (O-RADS) scores, and clinical variables.
Utilized internal and external test datasets for model validation.
Assessed model performance using Area Under the Curve (AUC) and false positive rates.
Evaluated the effect of OvcaFinder assistance on radiologist performance.

Main Results

OvcaFinder achieved AUCs of 0.978 (internal) and 0.947 (external), outperforming clinical and deep learning models.
Radiologist AUCs improved from 0.927 to 0.977 (internal) and 0.904 to 0.941 (external) with OvcaFinder assistance.
False positive rates decreased by 13.4% (internal) and 8.3% (external).
Inter-reader agreement among radiologists was enhanced.

Conclusions

OvcaFinder demonstrates superior performance in ovarian cancer diagnosis.
The interpretable AI model significantly improves diagnostic accuracy and consistency for radiologists.
OvcaFinder shows great potential to aid in the early and accurate detection of ovarian cancer.