Multimodal data integration using machine learning to predict the risk of clear cell renal cancer metastasis: A retrospective multicentre study

Question

1. What is the prognosis of ccRCC patients with metastasis?

2. What inclusion criteria were used in the research?

3. How many ccRCC patients developed metastases in the training set?

4. What was the protocol for enhanced CT scanning?

Accepted Answer

The prognosis of ccRCC patients with metastasis is unfavorable, with a median survival time of only 10.2 months. Approximately one-third of patients who underwent tumor removal surgery experienced recurrence and metastasis. The ability to estimate the likelihood of metastasis in patients with ccRCC remains a crucial issue that needs to be addressed in clinical diagnosis and treatment. Various clinical indicators, such as patient demographics, pathological stage, and laboratory test results, can serve as predictors of metastasis to some extent. However, these clinical factors alone fail to adequately account for the variation in clinical outcomes. Imaging studies, including contrast-enhanced computed tomography and ultrasonography, can provide further insight into the risk of metastasis. The integration of radiological and clinical features through fusion models may improve the prediction of metastasis risk in ccRCC patients, aiding clinicians in decision-making and tailoring treatment plans.

Accepted Answer

The research used two inclusion criteria: (i) patients with a confirmed diagnosis of ccRCC, and (ii) those with complete examination data, including contrast-enhanced CT scan images, ultrasound images, and laboratory data. Patients with inaccessible examination images, incomplete information, or a history of malignancy at other sites were excluded. The study enrolled 251 patients from three different sites.

Accepted Answer

In the training set, 50 ccRCC patients developed metastases. This set comprised 166 patients recruited from Shandong University Qilu Hospital (Jinan) between January 2015 and January 2023. The patients were divided into the training cohort, and the development of metastases was observed in 50 cases. This information is crucial for researchers studying the progression and treatment outcomes of ccRCC patients.

Accepted Answer

The protocol for enhanced CT scanning involved a tube voltage of 120KV, variable tube current settings between 200-350 mAs depending on the patient's body size, a slice thickness of 1.0 mm, a matrix of 512 x 512, and a pixel size of 0.625 x 0.625 mm. Contrast media, Ultravist and Omnipaque, were used at a rate of 2.5-3.5 mL per second and a volume of 60-80 mL per kilogram, proportional to body weight. Multiphase contrast-enhanced CT was performed in the cortical phase (30-35 seconds after contrast injection) and the medullary phase (70-75 seconds after contrast injection). The patient was asked to breathe smoothly during the imaging process.

Accepted Answer

In the study, the radiological characteristics analyzed included the maximum cross-sectional diameter of the lesion, the site of the lesion (left or right kidney), and the presence of signs such as hemorrhage, calcium, necrosis, lobulation, and encapsulation. Two radiologists with 12 and 10 years of experience in abdominal diagnosis, respectively, performed the analysis, and in case of dispute, a senior radiologist with 20 years of experience decided the outcome. The analysis was conducted using precontrast phase imaging, cortical phase, and medullary phase.

Accepted Answer

CT images were segmented by two experienced radiologists using ITK-SNAP v3.6.0. The software is designed for medical image segmentation. Radiologists scanned images in sections, selected the section with maximum cross-sectional area of the lesion, and manually drew a 2D region-of-interest (ROI) covering the entire axial section area of the lesion. Two experienced sonographers also delineated the ROI using ITK-SNAP following the same processing steps. This ensured accurate and consistent segmentation of the CT images for further analysis.

Accepted Answer

The Mann-Whitney U test was used for radiomics feature selection. This non-parametric test was employed to analyze all extracted radiomics features, and only those with a p-value less than 0.05 were retained. The test helps determine if there are statistically significant differences between two independent groups, in this case, the radiomics features. By using the Mann-Whitney U test, researchers can identify and retain features that show significant differences, ensuring the robustness of the radiomics analysis.

Accepted Answer

In model construction, selected clinical-imaging features were input into machine learning models such as logistic regression, support vector machine, naive Bayes, and XGBoost. The differences in model performance were compared to determine the best-performing model. A 5-fold cross-validation was utilized to obtain the final clinical model. The building process of the radiomic model was similar to the clinical model. After Lasso feature screening, the final selected features were inputted into the same machine learning models. Fair comparison was ensured by using 5-fold cross-validation and a test cohort. The efficiency of single modal and fusion modal (early fusion) was compared to evaluate the predictive performance of different models. Finally, a combined model was constructed by combining clinical-radiological characteristics and radiomics features. The predictive performance of the combined model was examined using testing sets and ROC curves. The clinical efficiency of the combined model was evaluated using decision curve analysis (DCA).

Accepted Answer

The mRMR method was employed for data processing to eliminate irrelevant and redundant features and select those with high consistency. Lasso regression analysis was conducted to further optimize the resulting subset. Finally, the number of features was decreased to 23. The selected 23 radiomics features (early fusion) were used to develop the combined model. The outstanding predictive ability of this model was evidenced by its AUC values of 0.870, 0.836, and 0.804 in the training set, test set 1, and test set 2, respectively, which were higher than the performance of the model according to a single-time image. These results confirmed our hypothesis that the predictive performance of the fused model was higher than that of the models constructed using a single source of information (plain scan, cortical phase, medullary phase, or ultrasound images).

Accepted Answer

The combined model, which fuses clinical-radiological characteristics and radiomics features, has shown improved AUC values in training, testing, and validation sets. Studies indicate that combining CT and ultrasound imaging features with critical clinical features significantly enhances prediction accuracy compared to clinical and radiological models alone. The combined model's DCA results demonstrate a substantial benefit in predicting the risk of lymph node metastasis (LNM) in ccRCC patients, with a corresponding prediction probability. Overall, the radiomic combined model has been shown to provide better clinical benefits in predicting LNM risk in ccRCC patients.

Multimodal data integration using machine learning to predict the risk of clear cell renal cancer metastasis: A retrospective multicentre study

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What is the prognosis of ccRCC patients with metastasis?

2. What inclusion criteria were used in the research?

3. How many ccRCC patients developed metastases in the training set?

4. What was the protocol for enhanced CT scanning?

5. What radiological characteristics were analyzed in the study?

6. How were CT images segmented in the study?

7. What statistical test was used for radiomics feature selection?

8. How were machine learning models compared in model construction?

9. What method was used for feature selection in radiomics models?

10. How does the combined model improve prediction accuracy?

References

Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries

Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012.

Radiomics: extracting more information from medical images using advanced feature analysis.

Renal-cell carcinoma.

European Association of Urology Guidelines on Renal Cell Carcinoma: The 2019 Update.

Related Papers (5)

Characterization of solid renal neoplasms using MRI-based quantitative radiomics features

The longest tumor diameter in one dimension as a predictor for skeletal metastasis in renal cell carcinoma.

Clear cell renal cell carcinoma: CT-based radiomics features for the prediction of Fuhrman grade

A CT-based deep learning model for predicting the nuclear grade of clear cell renal cell carcinoma

The radiomics-based tumor heterogeneity adds incremental value to the existing prognostic models for predicting outcome in localized clear cell renal cell carcinoma: a multicenter study