Kidney tumours are one of the most serious urological problems. They often grow without any noticeable symptoms in their early stages. Because they often don't show any symptoms, many cases go undetected until the disease has progressed to a more advanced stage, which makes treatment less effective and lowers survival rates. So, finding problems early and correctly is very important for improving patient outcomes. Computed Tomography (CT) imaging is commonly employed for kidney tumour diagnosis because it offers comprehensive cross-sectional representations of renal structures. But looking at CT scans by hand can take a long time and be different for each person. This study tackles these issues by suggesting an intelligent and automated system for finding kidney tumours that uses fuzzy image enhancement, deep learning, ensemble machine learning, and MLOps practices. The proposed method starts with a fuzzy inference system that is meant to improve CT images of the kidneys. Medical images frequently exhibit low contrast, noise, and inconsistent illumination, which may obscure subtle tumour regions. The fuzzy system looks at the distributions of pixel intensities and uses adaptive enhancement rules based on Lotfi A. Zadeh's fuzzy logic principles. The fuzzy-based method makes features more visible without losing detail or causing over-saturation, which is what happens with traditional contrast enhancement methods. This step of preprocessing makes sure that important structural details stand out before feature extraction. After the upgrade, the system uses two pre-trained deep convolutional neural networks: DenseNet121 and ResNet101. Gao Huang came up with the ideas for DenseNet, and Kaiming He came up with the ideas for ResNet. DenseNet121 uses dense connectivity patterns to make feature reuse easier and gradient flow better across layers. This lets it learn quickly even with small medical datasets. ResNet101, on the other hand, has residual connections that let very deep networks train well without running into problems with gradients that disappear. Both models use transfer learning from large image datasets to get high-level and discriminative features from enhanced CT images. We use a feature fusion strategy to combine the features taken from the two PT-DCNNs to make classification work even better. This ensemble representation takes information from both architectures that work well together, which gives us a feature set that is richer and more useful. We don't just use deep learning for classification. Instead, we use a weighted ensemble classifier that combines Support Vector Machines (SVM) and Random Forest (RF) to do the job. SVM helps with strong margin-based classification, and Random Forest makes things more stable by using multiple decision trees. A weighted averaging mechanism combines their predictions to make a final decision that is more stable and reliable. Using data augmentation methods, the dataset was made bigger to make it more generalisable and stable. We made different versions of CT images by adding controlled noise and fake distortions to make them look like real-world imaging problems. This process makes the model better able to work correctly in a variety of clinical situations. Also, using Machine Learning Operations (MLOps) practices makes sure that everything can be repeated, scaled up, and deployed smoothly in clinical settings. Model versioning, monitoring, and retraining tools help the system get better all the time. The experimental results show that the model works very well, with 99.2% accuracy on high-quality CT images and 98.5% accuracy on noisy images. These results are better than those of many traditional machine learning and deep learning methods that work on their own. The proposed automated system can help urologists and radiologists by giving them a reliable second opinion, lowering the number of diagnostic mistakes, making their jobs easier, and making it easier to act quickly. In the end, this smart framework is a big step forward for computer-aided medical diagnosis and could help patients live longer and make healthcare more efficient.
Jahan et al. (Wed,) studied this question.