Accurate and rapid diagnosis of kidney cancer from CT images plays a crucial role in timely clinical decision-making and reducing patient risk. In this study, a novel framework based on image preprocessing, edge extraction, local statistical feature analysis, and Deep Learning (DL) using a Recurrent Neural Network (RNN) is proposed for the detection and classification of kidney diseases. The proposed approach eliminates irrelevant image regions and extracts discriminative features through histogram analysis within multiple local windows and by investigating the relationships between image edges and different anatomical regions. These features effectively characterize the structural and textural properties of kidney tissues. To enhance class separability, adaptive histogram intervals are employed. Specifically, finer histogram bins are assigned to pixels with lower intensity values, while wider intervals are utilized for brighter pixels. This adaptive strategy enables the extraction of more informative statistical descriptors and improves the discrimination capability among different disease categories. Experimental results obtained on the standard Kaggle kidney CT image dataset demonstrate that the proposed method achieves a classification accuracy of 99.76% across four classes, namely normal, cyst, stone, and tumor. The obtained results indicate that the proposed framework outperforms previously reported methods and provides a highly effective solution for automated kidney disease classification.