Related Experiment Video
Updated: Jan 10, 2026
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
2.0K
Multimodal deep learning framework integrating multiphase CT and histopathological whole slide imaging for predicting
Changyi Ma1, Bao Feng2,3, Yan Lei1
1Department of Radiology, Jiangmen Central Hospital, 23 Beijie Haibang Street, Jiangmen, People's Republic of China.
Scientific Reports
|November 21, 2025
Summary
This study developed a deep learning model integrating CT scans and pathology images to predict clear cell renal cell carcinoma (ccRCC) recurrence after surgery. The CPNet model shows promise for improving postoperative risk stratification in ccRCC patients.
Area of Science:
- Oncology
- Radiology
- Pathology
- Artificial Intelligence
- Medical Imaging
Background:
- Clear cell renal cell carcinoma (ccRCC) is an aggressive cancer with a poor prognosis, necessitating accurate risk stratification.
- Current prognostic assessments often require multi-modal data, as radiological images have limitations and pathological images provide micro-level details.
- Integrating radiological and pathological data is crucial for improving the prediction of ccRCC outcomes.
Purpose of the Study:
- To develop and validate a deep learning (DL) fusion model using multiphase CT images and whole slide images (WSIs) for postoperative risk stratification in ccRCC patients.
- To compare the performance of different DL fusion models in predicting disease-free survival (DFS) after surgery.
- To evaluate the clinical utility and prognostic potential of the developed DL model.
Main Methods:
- A retrospective study included 274 ccRCC patients with multiphase CT scans and histopathological confirmation.
- The cohort was divided into training (164 patients) and testing (110 patients) sets.
- Deep learning models, including the CT and Pathology Mutual Guidance Fusion Diagnostic Network (CPNet), were developed and validated using CT images and WSIs to predict DFS.
- Model performance was assessed using accuracy (ACC), receiver operating characteristic (ROC) curve analysis, integrated discrimination improvement (IDI), and decision curve analysis (DCA).
Main Results:
- The PCP-Pathology Fuse model, a component of CPNet, achieved the highest AUC of 0.8363 and accuracy of 75.45%, outperforming other fusion models.
- The CPNet model demonstrated superior performance in predicting postoperative DFS in ccRCC patients.
- IDI and DCA confirmed significant net benefits for the PCP-based model, indicating its clinical utility.
- The model's performance was comparable to models using three-phase CT and pathology data.
Conclusions:
- The PCP-based CPNet model shows significant promise for predicting postoperative DFS in ccRCC patients.
- This DL fusion model can serve as a potential bioimaging prognostic marker for ccRCC.
- External validation is recommended to support the clinical integration of this prognostic tool.
