Kidney CT Scan Image Dataset for Cystic and Non-Cystic Classification
Description
This dataset contains a curated collection of Computed Tomography (CT) scan images designed for the development and benchmarking of deep learning models in medical image analysis. The dataset focuses on the binary classification of renal pathology, distinguishing between cystic structures and normal (non-cystic) kidney tissue. This collection serves as a valuable resource for researchers working on Computer-Aided Diagnosis (CAD) systems for nephrology. Data Acquisition The raw data was collected in direct collaboration with clinical specialists at various healthcare facilities. Each image was carefully reviewed and labeled by professional radiologists to ensure the accuracy of the ground truth. Cystic (Positive Class): 167 images Non-Cystic (Negative Class): 156 images Data Augmentation To enhance the generalization capability of convolutional neural networks (CNNs) and reduce overfitting, the dataset was expanded using 10 different augmentation techniques (e.g., rotation, horizontal and vertical flipping, zooming, shearing, and brightness normalization). Total Augmented Dataset: 3,852 images Cystic: 1,992 images Non-Cystic: 1,860 images Technical Specifications Domain: Medical Imaging / Deep Learning Modality: Computed Tomography (CT Scan) Task: Binary Classification (Cyst vs. Non-Cyst) Format: [PNG] Data Validation: Expert-labeled clinical data Usage and Ethics This dataset is provided for academic and research purposes. Users are expected to maintain ethical standards regarding the use of clinical data. Ensure all patient identifiers have been fully anonymized prior to any model training or publication of results. When using this data, appropriate citation of the original source is required.
Files
Steps to reproduce
To facilitate the reproduction of experiments and ensure the seamless integration of this dataset into machine learning workflows, please follow the structured pipeline outlined below: 1. Environment Configuration: Utilize a Python-based environment (Python 3.8+ recommended) with essential deep learning and image processing libraries installed. The recommended stack includes TensorFlow/Keras or PyTorch, NumPy, Pandas, OpenCV, and Scikit-learn. 2. Dataset Preprocessing: Normalization: Scale pixel intensity values to the range [0, 1] or normalize according to the requirements of the chosen pre-trained model. Resizing: Standardize all input images to a consistent resolution (e.g., 224 x 224 pixels) to maintain architectural compatibility. 3. Data Splitting: Partition the dataset into training, validation, and testing sets. A standard distribution of 70% for training, 15% for validation, and 15% for testing is recommended to ensure robust model evaluation. 4. Model Training: Loss Function: Utilize Binary Cross-Entropy as the objective function. Optimization: Implement the Adam or Stochastic Gradient Descent (SGD) optimizer with a starting learning rate of 10^-3, incorporating learning rate schedulers to improve convergence. Training Parameters: Configure batch size (e.g., 32) and training duration based on convergence monitoring on the validation set. 5. Evaluation Metrics: Assess the model's performance on the hold-out test set by calculating Accuracy, Precision, Recall, and F1-Score. The generation of a Confusion Matrix is highly encouraged to analyze the distribution of classification errors between the Cyst and Non-Cyst classes.
Institutions
- University of HalabjaSulaymaniyah, Halabja