Kidney Tumor Detection and Classification Using Convolutional Neural Network Architecture
Gurjot Kaur, Neha Vaishnavi Sharma, Sonal Jain Malhotra, Swati Devlival, Rupesh Gupta · 2024
This research focuses on developing a deep-learning model for the classification of medical images to detect tumors. The study utilizes a dataset consisting of 8,400 images labeled as either “Normal” or “Tumor.” The images are preprocessed by resizing to 256x256 pixels. The dataset is split into training, testing, and validation sets with ratios of 64%, 20%, and 16%, respectively. To balance the data, each class contains an equal number of images. The model architecture is based on a Sequential Convolutional Neural Network (CNN), incorporating layers such as Batch Normalization, Conv2D, MaxPooling2D, Dropout, Flatten, and Dense layers. The model is compiled with the Adam optimizer and Sparse Categorical Cross entropy loss function. Data augmentation techniques, including shear, zoom, and horizontal flip, are applied to the training set using the Image Data Generator. Training is conducted over 50 epochs, achieving a final training accuracy of 97.69% and validation accuracy of 95.31%. The model's performance is evaluated using various metrics, including accuracy, loss, and confusion matrix, which indicate high precision and recall values of 0.97 for both classes. Visualization of training and validation accuracy and loss over epochs demonstrates the model's effectiveness and stability. The research concludes that the proposed CNN model effectively classifies medical images for tumor detection, with potential applications in automated diagnostic systems. Future work could explore more advanced architectures and larger datasets to further enhance the model's performance and generalizability.