Preface
The long-awaited 39th Annual Meeting of the European Urological Association (EAU24) in 2024 was held in Paris, France from April 5 to 8 local time. As the largest and most influential urology conference in Europe, EAU24 has an extraordinary impact on the international urology community. This event not only attracted urology experts and scholars from all over the world, but also brought together the latest research results and cutting-edge technologies in the field of urology, injecting new vitality into the development of the field of urology.
At the "Progress in the Diagnosis, Intervention and Active Monitoring of Small Renal Masses" held today, Professor Xiong Ying from the research team of Professor Guo Jianming from the Department of Urology of Zhongshan Hospital Affiliated to Fudan University reported a study on "Artificial Intelligence Using Real-world Data to Predict the Pathological Features and Clinical Outcomes of Kidney Masses through Preoperative Multi-period CT Images" (Abstract No. A0458). With its forward-looking and innovative nature, the study was selected as the best abstract in the field of kidney cancer by this EAU24, bringing a new perspective to the preoperative diagnosis of small renal masses and sending a strong voice in China. Yimaitong is hereby compiled below for the benefit of readers.
Expert Profile
Prof. Guo Jianming
Zhongshan Hospital Affiliated to Fudan University
- Director of the Department of Urology, Zhongshan Hospital, Fudan University, Professor, Doctoral Supervisor
- Member of the Standing Committee of the Urology Branch of the Chinese Medical Doctor Association
- Member of the Oncology Group of the Urology Branch of the Chinese Medical Association
- Vice Chairman of the Urology Branch of Shanghai Medical Association and Head of the Oncology Group
- Vice Chairman of the Urological Oncology Committee of Shanghai Anti-Cancer Association
- Vice Chairman of Shanghai Urology and Andrology Branch of Integrated Traditional Chinese and Western Medicine,
- Member of the Standing Committee of the Urological Oncology Committee of the Chinese Anti-Cancer Association
- Member of the Standing Committee of the Male Germline Tumor Committee of the Chinese Anti-Cancer Association
- Member of the Standing Committee of the Prostate Cancer and Urothelial Carcinoma Committee of the Chinese Clinical Oncology Association
- Vice Chairman of the Urology Branch of the Chinese Society of Sexology
- Member of the Standing Committee and Deputy Secretary-General of the World Association of Chinese Urologists
- Vice Chairman of the Urology Branch of the Cross-Strait Medical Exchange Association
- Vice Chairman of the Urology and Andrology Committee of the Chinese Geriatric Health Care Association
Expert Profile
Professor Xiong Ying
Zhongshan Hospital Affiliated to Fudan University
- Attending physician of the Department of Urology, Zhongshan Hospital, Fudan University
- Under the supervision of Professor Guo Jianming, he mainly focuses on the use of pathology, imaging and multi-omics data to realize AI-assisted intelligent diagnosis and treatment of kidney cancer. In recent years, he has published 10 SCI papers in well-known oncology journals such as Radiology and J Immunother Cancer as the first author and co-first author. He presided over a project of the National Natural Science Foundation of China. He was selected into the Harvard Global Clinical Scholar Training GCSRT Program and the Outstanding Young Talents Program of Zhongshan Hospital. He won the overall champion of the Eastern Division of the 2023 Exploration Talent Show Urology Young Physician Surgery & Case Challenge, and the 2018 Shanghai Outstanding Graduate.
A0458: Artificial intelligence uses real-world data to predict the pathological characteristics and clinical outcome of renal masses through preoperative multi-period CT images
1. Background
With the popularization of imaging examination techniques such as CT and MRI, the detection rate of asymptomatic kidney tumors is increasing year by year. For such incidentally discovered renal masses, treatment decisions are usually made in the context of unclear pathology. Therefore, improving the diagnostic accuracy of benign and malignant renal masses and further distinguishing the degree of malignancy (invasive/indolent) of renal masses are essential for the selection of subsequent treatment regimens.
2. Study design
The research team analyzed 13,261 preoperative CT images of 4,557 patients from six hospitals in China, including Zhongshan Hospital Affiliated to Fudan University, the First People's Hospital Affiliated to Zhejiang University, Zhangye People's Hospital, Quanzhou People's Hospital, and Xiamen Hospital of Zhongshan Hospital Affiliated to Fudan University, and five public tumor imaging databases. Kaplan-Meir analysis and Cox regression analysis were used to compare the difference in survival between AI-predicted indolent and aggressive tumors. Assessment of genomics, transcriptomics, and immune landscape using bioinformatics analysis and immunohistochemistry.
3 Research results
The benign and malignant diagnostic model had better predictive performance than the current radiomics prediction model and RENAL score prediction model in each validation cohort, and the AUC value reached 0.853~0.898 in different validation cohorts, and its diagnostic accuracy exceeded the average performance of seven experienced radiologists in the prospective validation cohort. In addition, AI diagnostic models can help radiologists significantly improve the accuracy of diagnosis.
The deep learning model that distinguishes between aggressive tumors and indolent tumors also outperformed the current radiomics model and the RENAL score prediction model in each validation cohort, and the AUC value reached 0.763~0.792 in different validation cohorts. In the external validation cohort, the survival of AI-predicted invasive kidney cancer was significantly shorter than that of AI-predicted indolent kidney cancer [disease-specific survival (DSS), p<0.001, HR=20.81; recurrence-free survival (RFS), p<0.001, HR=9.71; overall survival (OS), p.]. <0.001，HR=13.27]。 The aggressiveness probability score determined by AI can be used as an independent adverse prognostic factor for patient survival outcomes, and its predictive power on clinical outcome is higher than that of postoperative pathological indicators such as TNM stage and WHO/ISUP grading system. Finally, bioinformatics analysis showed that aggressive renal cancer was in an immunosuppressive microenvironment with higher CD8+ T cell and regulatory T cell infiltration than indolent renal cancer.
4. Conclusions of the study
The deep learning model can non-invasively predict the likelihood of malignant and invasive lesions of a kidney mass based on preoperative multiphasic CT images. In addition, AI-predicted aggressive cancers are associated with poorer survival outcomes.
Professor Guo Jianming commented
This study aims to address the complexity and differences in the imaging diagnosis of kidney tumors in different regions. CT imaging plays a central role in the diagnosis of kidney tumors, however, its application also faces two major challenges. First, in some atypical masses, it is sometimes difficult to distinguish benign from malignant, and about 20% of atypical masses are difficult to judge in nature. If misjudged, patients with benign tumors will suffer unnecessary losses. Therefore, it is of great significance to improve the accuracy of benign and malignant differentiation to protect patients' renal function and reduce unnecessary surgical intervention. Secondly, how to accurately assess the degree of invasion and prognosis of malignant tumors with the help of imaging methods is also an urgent problem to be solved.
In order to overcome these problems, our research team cooperated with six hospitals to include more than 13,261 CT imaging data from 4,557 patients and carried out in-depth research to successfully construct this deep learning model. Excitingly, the predictive model has shown superior performance across validation cohorts, demonstrating the great potential of AI in the imaging diagnosis of kidney tumors. This discovery has important implications for us, and the research team will continue to expand the scope of research in the future, with a view to promoting the continuous advancement of imaging diagnostic technology for kidney tumors. It is hoped that this research result can help doctors in more hospitals to achieve accuracy in the diagnosis of kidney tumor images, and provide patients with more scientific and reasonable treatment options, especially for hospitals in the central and western regions or with less experience, AI prediction models will provide strong technical support and promote the widespread popularization and improvement of kidney tumor diagnosis technology.
编辑:Gardenia审校:郭剑明教授执行:Gardenia
This platform aims to deliver more medical information to healthcare professionals. The content published on this platform should not be used as a substitute for professional medical guidance in any way, nor should it be regarded as diagnosis and treatment advice. If such information is used for purposes other than understanding medical information, the platform does not assume relevant responsibilities. This platform does not mean that it agrees with its descriptions and views on the published content. If copyright issues are involved, please contact us, and we will deal with it as soon as possible.
Yimaitong is a professional online doctor platform, and the mission of the platform is to "sense the pulse of the world's medicine and help China's clinical decision-making". Yimaitong has a series of products such as "Clinical Guidelines", "Medication Reference", "Medical Literature King", "Yizhiyuan", "eYantong" and "ePulse", which fully meet the needs of medical workers in clinical decision-making, obtaining new knowledge and improving scientific research efficiency.