Enhanced deep learning-based decision support system for kidney tumor detection

Abstract

Accurate and early detection of kidney tumors remain a significant challenge in clinical radiology, particularly due to the subtle presentation of tumors in early stages and variability in human interpretation. This study proposes a deep learning–based decision support system for automated classification of kidney tumors using computed tomography (CT) images. A publicly available dataset of 10,000 grayscale CT scans—equally representing healthy and tumor cases—was processed through normalization, resizing, and filtering of non-horizontal views. Five convolutional neural networks (CNNs) were evaluated: AlexNet, EfficientNet-B0, Xception, Darknet-53, and DenseNet-201. DenseNet-201 achieved the best performance, with an accuracy of 96.20%, precision of 1.0000, and recall of 92.22%. Evaluation metrics were derived from confusion matrices, and the influence of learning rate on model performance was examined. Compared to related methods trained on smaller datasets, the proposed system demonstrated strong generalizability and competitive accuracy, indicating its potential utility in clinical decision support for kidney cancer diagnosis.

Article Content

Keywords

Article Info