KIDNEY HYDRONEPHROSIS STAGE CLASSIFICATION USING DEEP LEARNING AND IMAGE PROCESSING

Abstract

Kidney hydronephrosis is a common condition characterized by the dilation of the renal pelvis and calyces due to obstruction in the urinary tract. In, today, where multiple scanned images are obtained for each patient, the current hydronephrosis stage classification is time-consuming, depends on the knowledge and the mood of the radiologist, and is susceptible to variability in classification between and among observers. Deep learning techniques can be utilized for kidney hydronephrosis stage classification. Utilizing deep learning in kidney hydronephrosis stage classification reduces the workload of radiologists and increases diagnosis accuracy. In this thesis, we explore the use of deep learning models, specifically VGG16, ResNet50, and InceptionV3, combined with image preprocessing techniques like noise reduction, image resize, augmentation, segmentation, feature extraction, and image normalization to improve model performance. Additionally, we conduct comparative analyses to assess the robustness and generalization capabilities of the fine-tuned models across kidney CT-scanned datasets. A total of 7400 CT scanned images are used to build the model with 80% of the dataset used for training, 10 % for validation, and 10% for testing. Features from kidney CT images are extracted using transfer learning with pre-trained VGG16, ResNet50, and InceptionV3 models and fed into fully connected layers for stage classification. The performance of each model is evaluated using model evaluation metrics of accuracy, f1 score, and recall. The accuracy score of ResNet50, VGG16, and InceptionV3 is 79%,91%, and 74% respectively, 87%,86% and,63% recall, 85%,86%, and 67% f1 scores. Comparing those values VGG16 has higher accuracy, recall, and f1 score, and is larger and is the best model for kidney hydronephrosis stage classification from CT scanned images. Misclassifications occurred due to the similar structure of kidney images at each level that caused hydronephrosis. Future work should address these challenges. This research contributes to the advancement of automated kidney hydronephrosis stage classification by using deep learning and image processing techniques. The findings highlight the potential of VGG16 as a powerful model for accurate and efficient classification of kidney hydronephrosis stages.