Adaptive dual-window enhancement and multi-scale texture prior fusion for robust kidney CT classification

Abstract

Accurate classification of kidney diseases is of great importance for clinical diagnosis and treatment. However, traditional CT images suffer from insufficient contrast, blurred tissue boundaries, and complex texture variations, which limit the performance of automated analysis. This paper proposes a novel kidney CT classification framework that combines Adaptive Dual-Window Enhancement (ADWE) with Multi-Scale Texture Prior Fusion (MTPF). The ADWE module dynamically adjusts window width and window level to generate complementary views, effectively enhancing the contrast of both soft tissues and high-density structures; the MTPF module incorporates edge, local binary pattern (LBP), and Gabor texture priors to achieve fine-grained structural modeling. Experimental results demonstrate that in the binary classification task, the proposed method achieves an accuracy of 0.9802, F1-score of 0.9786, and AUC of 0.9989, all outperforming mainstream deep learning and domain-specific medical models. In the four-class classification task, it achieves an accuracy of 0.8821, F1-score of 0.8438, and AUC of 0.9801, representing an improvement of approximately 3%–5% compared with the ConvNeXtV2 baseline. Moreover, under noise intensity σ = 0.1 , the method still maintains an accuracy of 0.8510 and an AUC of 0.9634, showing remarkable robustness. These results validate the effectiveness and clinical potential of the proposed method for automated kidney CT classification.