Model 1 · Clinical Risk Predictor
Trained on 36,738 SEER patients. Predicts overall metastasis risk and site-specific risks for lung, bone, liver, and brain.
Patient Demographics
Tumour Staging
Histopathology
Model 2 · Genomic Transcriptomic Predictor
Trained on 418 TCGA-KIRC RNA-Seq patients. Uses 54 variance-filtered genes selected via ANOVA F-test. Achieved 94.4% Recall.
gene, expression_value_TPM
One gene per row. Standard output from featureCounts, STAR, Salmon.
Enter normalized RNA-Seq expression values (RSEM/TPM) for each gene. Typical range 0–20. Use the Demo button to auto-fill realistic high-risk values.
Model 3 · 3D Radiomic Imaging Predictor
Trained on TCGA-KIRC preoperative CT/MRI scans. Uses 49 PyRadiomics shape, intensity & texture features extracted from tumour segmentation.
Pre-operative NIfTI (.nii.gz)
TotalSegmentator v2
PyRadiomics runs server-side
XGBoost Model 3
feature, value
Each row is one PyRadiomics feature.
Enter the 49 radiomic features extracted from your tumour segmentation. Use the Demo button to load real values from a TCGA-KIRC patient with confirmed metastasis.
3-Modality Late Fusion · Final Prediction
Combines all three modalities via Bayesian Evidence, Dempster-Shafer, and Optimal Transport fusion. This is the final, publishable framework achieving AUROC 0.979.
Patient Demographics
Tumour Staging
Histopathology
Gene Expression Log2(FPKM+1)
🤖 2-Stage AI Imaging Pipeline
Upload a raw 3D CT scan (.nii or .nii.gz).
- Stage 1 (Validator): EfficientNet-B0 verifies the image is a Kidney CT.
- Stage 2 (Segmenter): AI automatically segments the tumour and extracts 49 PyRadiomics features.
Professional 3D DICOM Viewer
Launch the full-screen radiological workstation to inspect the CT scan and auto-segmentations in high resolution.
Manual PyRadiomics Features (49 Shape/Texture Features) — use Demo to auto-fill