Identification of renal cell carcinoma subtypes and clinical stages based on hyperspectral imaging technology

• Proposes RCTnet, a novel TCN-LSSVM hybrid model for hyperspectral renal tissue classification, enhancing spectral-temporal feature fusion and discriminative performance. • Achieves robust clinical validation (AUC >0.9) in multi-class RCC subtype and TNM stage classification using prospectively acquired data. • Increases diagnostic reproducibility and provides interpretable spectral insights to support personalized therapeutic strategies. • Delivers an instrumentation-ready computational pipeline advancing HSI as a quantitative tool for tissue phenotyping in clinical settings. Accurate classification of renal cell carcinoma (RCC) subtypes and clinical stages is crucial for treatment planning but remains challenging using conventional imaging and pathology alone. This study introduces RCTnet, a hybrid framework combining temporal convolutional networks (TCN) for sequential spectral feature extraction with least-squares support vector machines (LSSVM) for nonlinear classification. Hyperspectral imaging was applied to 230 RCC tissue specimens, including clear cell, papillary, and chromophobe RCC, renal oncocytoma, and TNM-staged ccRCC samples. Preprocessing with Savitzky–Golay filtering, derivative analysis, and principal component analysis ensured high-quality spectral inputs. RCTnet outperformed traditional SVM, basic CNN, ResNet50, and Vision Transformer baselines. It achieved 91.7% accuracy, 91.8% sensitivity, and 97.1% specificity for subtype classification, and 90.4% accuracy, 89.7% sensitivity, and 94.2% specificity for stage classification. These results highlight the synergy between TCN and LSSVM, as well as the physiological relevance of key spectral bands. RCTnet demonstrates strong potential for improving precision diagnosis and clinical decision-making in RCC.