Fuzzy Enhanced Kidney Tumor Detection: Integrating Machine Learning Operations for a Fusion of Twin Transferable Network and Weighted Ensemble Machine Learning Classifier

Kidney tumors, often asymptomatic, can lead to serious health problems if left undiagnosed. This study tackles the crucial issue of kidney tumor detection using CT scans. The proposed approach leverages the power of image enhancement using fuzzy systems, deep learning, and machine learning for automated kidney tumor detection in CT images. The study proposes a fuzzy inference system to enhance kidney CT image contrast. This system analyzes image data and uses fuzzy logic to adjust pixel intensities, aiming to improve the distinction between features in the image without creating over-enhancement. Two pre-trained deep convolutional neural networks (PT-DCNNs), DenseNet121 and ResNet101, are used to extract features from the enhanced CT images. These features capture essential characteristics that differentiate between normal and tumor-containing scans. Combining features from twin PT-DCNNs (ensemble approach) creates a richer representation of the image content. The informative features are fed into a combined classifier where Support Vector Machines and Random Forests are combined using a weighted average to achieve the final and potentially more robust classification of kidney tumors. To improve training, we amplified the original dataset by creating variations with added noise and artificial modifications to simulate real-world image imperfections. The integration of Machine Learning Operations practices ensures the scalability, reproducibility, and clinical deployment of the system. The model achieved an impressive accuracy of 99.2% on high-quality images and 98.5% on noisy images, surpassing traditional methods. This automated approach can assist urologists in confirming the presence of kidney tumors, minimizing human error during physical inspection and potentially leading to improved patient outcomes.