Published Paper Details:

Tumours of the kidney are among the most serious problems of urology. They often develop without any obvious symptoms in their early stages. Symptoms are often absent, so many cases are not diagnosed until the disease has progressed to a later stage, when treatment is less effective and survival rates are lower. Therefore, early and accurate problem identification is critical for improving patient outcomes. Computed Tomography (CT) imaging is often used in the diagnosis of kidney tumours, as it provides detailed cross-sections of renal structures. But manually reviewing CT scans is slow and subjective. This study addresses such issues by introducing an intelligent and automated system for the detection of kidney tumours, integrating fuzzy image enhancement, deep learning, ensemble machine learning, and MLOps practices. The proposed method is based on a fuzzy inference system for the improvement of the CT images of the kidneys. Medical images often suffer from low contrast, noise and inconsistent illumination that may obscure subtle tumour regions. The fuzzy system analyses the distributions of pixel intensity and uses adaptive enhancement rules based on the fuzzy logic principles of Lotfi . A. Zadeh. The fuzzy based method improves the features to be more prominent without losing the detail and over-saturation as in the case of traditional contrast enhancement methods. This preprocessing step highlights important structural details before feature extraction. Following the upgrade, the system employs two pre-trained deep convolutional neural networks, DenseNet121 and ResNet101. The ideas for DenseNet came from Gao Huang and the ideas for ResNet came from Kaiming He. DenseNet121 uses dense connectivity patterns to enable feature reuse and gradient flow across layers. This allows the fast learning even on small medical datasets. On the other hand, ResNet101 has residual connections that allow very deep networks to train well without running into problems with gradients that vanish. Both models utilise transfer learning on large image datasets to extract high-level, discriminative features from augmented CT scans. We use the feature fusion strategy to combine the features taken from the two PT-DCNNs to make the classification work even better. This combined representation uses information from both architectures that complement each other, providing us with a richer and more useful feature set. We don't just use deep learning for classification. Instead we use a weighted ensemble classifier combining Support Vector Machines (SVM) and Random Forest (RF) to do the job. SVM helps with classification with a strong margin. Random Forest makes things more stable using multiple decision trees. Their predictions are combined through a weighted averaging mechanism to make a final decision that is more stable and reliable. To increase the dataset and make it more generalisable and stable, data augmentation techniques were applied. We ran various versions of the CT images, adding controlled noise and fake distortions to simulate problems encountered in the real world of imaging. This process makes the model more capable to work properly in different clinical settings. Moreover, the application of Machine Learning Operations (MLOps) techniques enables to make all the processes reproducible, scalable and easily deployable in clinical practice. System is getting better all the time with model versioning, monitoring and retraining tools The experimental results show that the model performs excellently with accuracy of 99.2% on high quality CT images and 98.5% on noisy images. These results are better than many traditional machine learning and deep learning methods that work alone. The suggested automated system can assist urologists and radiologists by providing them with a reliable second opinion, reducing the number of diagnostic errors, making their work easier and allowing prompt action. Ultimately, this intelligent framework represents a major step forward in computer-assisted medical diagnosis and could help patients live longer and health care become more efficient.