Related Experiment Videos
Multimodal deep learning model for multiclass classification of renal tumors.
Shiwei Luo1, Quan Quan2,3,4, Ruimeng Yang5
1Department of Radiology, the Second Xiangya Hospital of Central South University, Changsha, Hunan, China.
NPJ Digital Medicine
|May 4, 2026
Summary
A new deep learning model, MPANet, accurately classifies common renal tumors using CT scans and clinical data. This AI tool shows significant potential to improve diagnostic accuracy and aid clinical decision-making for kidney cancer.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate pre-treatment classification of renal masses is vital for effective therapeutic strategies and improved patient outcomes.
- Distinguishing between common and confusable renal tumors like clear cell renal cell carcinoma (ccRCC), papillary renal cell carcinoma (pRCC), oncocytic neoplasms, and fat-poor angiomyolipoma (fpAML) remains a clinical challenge.
Purpose of the Study:
- To develop and validate the Multi-Phase Attention Network (MPANet), a multimodal deep learning model for the multiclass classification of four common renal tumors.
- To assess MPANet's performance using both complete and missing-phase contrast-enhanced CT data, integrated with clinical information.
Main Methods:
- Development of MPANet, a deep learning model integrating multiphase contrast-enhanced CT and clinical data.
- Training and validation on a multi-center dataset of 1688 cases.
- Comparison of MPANet's performance against assessments by radiologists using CT and MRI with the clear cell likelihood score (ccLS) system.
Main Results:
- MPANet consistently outperformed single-phase models across all test sets.
- In internal testing, MPANet achieved a macro-average AUC of 0.850 and accuracy of 73.3%, significantly exceeding radiologist performance (43.6-62.4%).
- External test sets showed strong performance with macro-average AUCs of 0.811 and 0.813.
Conclusions:
- MPANet demonstrates superior performance in classifying common renal tumors compared to human expert assessment.
- The model's ability to utilize incomplete CT data enhances its clinical applicability.
- MPANet holds significant potential as a clinical decision-support tool for personalized renal tumor diagnosis.
