Predicting postoperative prognosis in clear cell renal cell carcinoma using a multiphase CT-based deep learning
Changyin Yao1,2, Bao Feng3, Shurong Li4
1Department of Radiology, Jiangmen Central Hospital, Jiangmen, China.
Abdominal Radiology (New York)
|September 23, 2024
Summary
A new deep learning (DL) model using CT scans shows promise for predicting clear cell renal cell carcinoma (ccRCC) patient prognosis. Combining this AI model with the Leibovich score further enhances its predictive accuracy for better patient outcomes.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Existing clinicopathological risk stratification systems (CRSSs) like the Leibovich score aid in predicting clear cell renal cell carcinoma (ccRCC) prognosis.
- However, reliable noninvasive preoperative indicators for predicting postoperative prognosis in clinical practice are lacking.
Purpose of the Study:
- To evaluate the efficacy of a deep learning (DL) model, utilizing CT images, in predicting the postoperative prognosis of ccRCC patients.
Main Methods:
- A cohort of 382 ccRCC patients was retrospectively analyzed, divided into training (n=229) and testing (n=153) groups.
- ResNet50 was employed for feature extraction from precontrast, corticomedullary, and nephrographic phase CT images.
- Extreme learning machines (ELMs) constructed classification models, which were compared and combined with the Leibovich score.
Main Results:
- The three-phase CT-based DL model demonstrated superior performance (AUC=0.839) compared to single-phase models.
- A combined nomogram integrating the three-phase DL model and Leibovich score achieved the highest AUC (0.888).
- This combined approach significantly improved predictive performance (IDI > 0.13, P < 0.001).
Conclusions:
- A CT-based DL model shows potential as a valuable tool for preoperative prognosis prediction in ccRCC.
- Integrating the DL model with the Leibovich score can substantially enhance predictive accuracy, aiding clinical decision-making.
