AI-Driven Robust Kidney and Renal Mass Segmentation and Classification on 3D CT Images

Area of Science:

• Medical Imaging and Artificial Intelligence
• Oncology and Urology

Background:

• Early kidney cancer intervention improves survival rates.
• Manual segmentation of renal masses via CT scans is labor-intensive.
• Artificial intelligence (AI) offers potential for automated cancer diagnosis.

Purpose of the Study:

• To develop an AI-driven framework for automatic kidney and renal mass diagnosis.
• To identify abnormal kidney areas and predict renal cell carcinoma (RCC) histological subtypes.
• To reduce manual segmentation workload and avoid unnecessary invasive procedures.

Main Methods:

• An end-to-end AI framework utilizing a 3D deep learning architecture (Res-UNet) for kidney and renal mass segmentation.
• A dual-path classification network integrating local and global features for RCC subtype prediction (clear cell, chromophobe, oncocytoma, papillary, others).
• A weakly supervised learning schema to address domain gaps across different CT scanner vendors using minimal annotations.

Main Results:

• The AI framework accurately segmented kidney and renal mass regions.
• The system achieved high accuracy in predicting common RCC subtypes.
• The proposed system outperformed existing methods on the KiTs19 dataset, demonstrating robustness via cross-dataset validation.

Conclusions:

• The novel AI framework effectively automates kidney mass segmentation and RCC subtype diagnosis.
• Weakly supervised learning enhances the system's robustness across multi-institutional datasets.
• This AI approach has the potential to improve diagnostic efficiency and patient outcomes in kidney cancer care.