An accurate and interpreountrytable classification of kidney CT images is essential for the early diagnosis of renal diseases such as cysts, stones, and tumours, as a delayed or unreliable interpretation can lead to severe clinical complications. However, most existing deep learning models rely on single-scale feature extraction and offer limited explainability, thus limiting their clinical applicability. To address these limitations, this study proposes EagleVisionNet. This biologically inspired dual-path convolutional neural network explicitly models peripheral and foveal vision mechanisms to enable structured multi-scale feature learning. The peripheral pathway captures the global anatomical context, while the foveal pathway focuses on high-resolution localised pathological details. The model was trained and evaluated on a dataset of 12,446 CT images collected from multiple hospitals in Dhaka, Bangladesh, comprising four classes: Normal, Cyst, Stone and Tumor. Experimental results demonstrate an overall classification accuracy of 94%, with macro-averaged precision, recall, and F1-score of 0.93. Visual explanations based on Grad-CAM confirm that the proposed model consistently attends to clinically relevant kidney regions, thus enhancing interpretability and diagnostic trust. Multi-class ROC and Precision–Recall analyses further validate the robustness of the framework, particularly for imbalanced stone cases. These findings highlight the effectiveness of bio-inspired multi-scale architectures combined with explainable AI for developing reliable and clinically meaningful automated kidney disease diagnostic systems.