Shuanbao Yu , Ruochen Hao , Jinshan Cui , Yang Yang , Zeyuan Wang , Haoke Zheng , Minglin Li , Xinfei Li , Wei Chen , Wendong Jia , Meng Wang , Bo Chen , Pokhrel Gaurab , Yuhong Li , Dapeng Wu , Xuesong Li , Xuepei Zhang
{"title":"Integration of radiomic and deep features to reliably differentiate benign renal lesions from renal cell carcinoma","authors":"Shuanbao Yu , Ruochen Hao , Jinshan Cui , Yang Yang , Zeyuan Wang , Haoke Zheng , Minglin Li , Xinfei Li , Wei Chen , Wendong Jia , Meng Wang , Bo Chen , Pokhrel Gaurab , Yuhong Li , Dapeng Wu , Xuesong Li , Xuepei Zhang","doi":"10.1016/j.ejrad.2025.111989","DOIUrl":null,"url":null,"abstract":"<div><h3>Purpose</h3><div>Accurate differentiation of benign renal lesions from renal cell carcinoma (RCC) is crucial for optimized management, particularly for small renal lesions (≤4 cm in diameter). This study aimed to integrate clinical data, radiomic features, and deep features to improve diagnostic performance in distinguishing benign renal lesions from RCC.</div></div><div><h3>Methods</h3><div>In this retrospective study, radiomic and deep scores were constructed using the AutoGluon-Tabular classifier based on radiomic and deep features extracted from the whole region of CT imaging. These scores were combined with clinical score, derived from clinical data, to create a multimodal score. The multimodal score was developed using a training cohort of 1,030 patients, validated in a cohort of 448 patients, and tested in a cohort of 680 patients from two independent institutions.</div></div><div><h3>Results</h3><div>A total of 2,158 patients with renal lesions were evaluated, including 1,739 (80.6 %) diagnosed with RCC, 419 (19.4 %) with benign renal lesions, and 1,340 (62.1 %) with small renal lesions. The multimodal score demonstrated higher diagnostic performance for both overall renal lesions (AUC: 0.949–0.906 vs 0.926–0.765) and small renal lesions (AUC: 0.929–0.880 vs 0.901–0.657) compared to the clinical score, radiomic score, deep score, and radiologists’ interpretations in the validation and test cohorts. The multimodal score also showed satisfactory concordance and the highest net benefit across threshold probabilities exceeding 60 %. The AUC values for subgroup with predicted risk below or above cutoff values for 99 % sensitivity or specificity were 0.964 and 0.897 for overall and small renal lesions in the external test cohort, respectively.</div></div><div><h3>Conclusions</h3><div>The multimodal score is a reliable predictor for distinguishing benign renal lesions from RCC in both overall and small renal lesions, and has the potential to spare patients from unnecessary biopsies or surgeries and can aid in guiding the management for renal lesions.</div></div>","PeriodicalId":12063,"journal":{"name":"European Journal of Radiology","volume":"184 ","pages":"Article 111989"},"PeriodicalIF":3.2000,"publicationDate":"2025-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"European Journal of Radiology","FirstCategoryId":"3","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0720048X25000750","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"RADIOLOGY, NUCLEAR MEDICINE & MEDICAL IMAGING","Score":null,"Total":0}
引用次数: 0
Abstract
Purpose
Accurate differentiation of benign renal lesions from renal cell carcinoma (RCC) is crucial for optimized management, particularly for small renal lesions (≤4 cm in diameter). This study aimed to integrate clinical data, radiomic features, and deep features to improve diagnostic performance in distinguishing benign renal lesions from RCC.
Methods
In this retrospective study, radiomic and deep scores were constructed using the AutoGluon-Tabular classifier based on radiomic and deep features extracted from the whole region of CT imaging. These scores were combined with clinical score, derived from clinical data, to create a multimodal score. The multimodal score was developed using a training cohort of 1,030 patients, validated in a cohort of 448 patients, and tested in a cohort of 680 patients from two independent institutions.
Results
A total of 2,158 patients with renal lesions were evaluated, including 1,739 (80.6 %) diagnosed with RCC, 419 (19.4 %) with benign renal lesions, and 1,340 (62.1 %) with small renal lesions. The multimodal score demonstrated higher diagnostic performance for both overall renal lesions (AUC: 0.949–0.906 vs 0.926–0.765) and small renal lesions (AUC: 0.929–0.880 vs 0.901–0.657) compared to the clinical score, radiomic score, deep score, and radiologists’ interpretations in the validation and test cohorts. The multimodal score also showed satisfactory concordance and the highest net benefit across threshold probabilities exceeding 60 %. The AUC values for subgroup with predicted risk below or above cutoff values for 99 % sensitivity or specificity were 0.964 and 0.897 for overall and small renal lesions in the external test cohort, respectively.
Conclusions
The multimodal score is a reliable predictor for distinguishing benign renal lesions from RCC in both overall and small renal lesions, and has the potential to spare patients from unnecessary biopsies or surgeries and can aid in guiding the management for renal lesions.
期刊介绍:
European Journal of Radiology is an international journal which aims to communicate to its readers, state-of-the-art information on imaging developments in the form of high quality original research articles and timely reviews on current developments in the field.
Its audience includes clinicians at all levels of training including radiology trainees, newly qualified imaging specialists and the experienced radiologist. Its aim is to inform efficient, appropriate and evidence-based imaging practice to the benefit of patients worldwide.