AISSLab Breast Cancer Sub-datasets: Task-Oriented Structuring of Mammographic Images for Classification Scenarios
Description
The AISSLab Breast Cancer Subdatasets provide a structured and task-oriented collection of mammographic images designed for AI-based classification research. The dataset is organized into Raw and Augmented subdatasets, enabling effective training and evaluation of deep learning models. Sub-datasets overview: AISSLab-v1: Normal (100) vs. Abnormal (166) → 266 raw images → 1064 augmented images AISSLab-v2: Normal (100) vs. Malignant (100) → 200 raw images → 800 augmented images AISSLab-v3: Benign (66) vs. Malignant (100) → 166 raw images → 664 augmented images AISSLab-v4: Benign (66), Malignant (100), Normal (100) → 266 raw images → 1064 augmented images AISSLab-v5: BI-RADS 2 (18), BI-RADS 3 (48), BI-RADS 4 (54), BI-RADS 5 (46) → 166 raw images → 664 augmented images This modular design allows researchers to experiment with binary and multiclass classification tasks, while the applied data augmentation helps increase sample size and reduce class imbalance. The dataset complies with the FAIR Data Principles to ensure reusability, transparency, and interoperability. This version transforms AISSLab into a modular and reusable benchmark for medical AI researchers working on classification tasks in mammography. Citation Researchers utilizing the AISSLab Subdatasets are kindly requested to cite the following works: A. M. Al-Hejri, R. M. Al-Tam, M. Fazea, A. H. Sable, S. Lee, and M. A. Al-antari, “ETECADx: Ensemble Self-Attention Transformer Encoder for Breast Cancer Diagnosis Using Full-Field Digital X-ray Breast Images,” Diagnostics, vol. 13, no. 1, Jan. 2023, doi: 10.3390/diagnostics13010089. R. M. Al-Tam, A. M. Al-Hejri, S. S. Alshamrani, M. A. Al-antari, and S. M. Narangale, “Multimodal breast cancer hybrid explainable computer-aided diagnosis using medical mammograms and ultrasound Images,” Biocybern Biomed Eng, vol. 44, no. 3, pp. 731–758, Jul. 2024, doi: 10.1016/j.bbe.2024.08.007. Contact For any inquiries regarding the dataset, please feel free to reach out via email: tugcecoban74@gmail.com
Files
Steps to reproduce
1. Access the Dataset Download the AISSLab Breast Cancer Subdatasets from Mendeley Data. 2. Understand the Dataset Structure Refer to the AISSLab_Subdataset_README.txt file located at the root directory for detailed descriptions of each subdataset (AISSLab-v1 to AISSLab-v5). Each subdataset corresponds to a specific classification task (e.g., Normal vs. Abnormal, Benign vs. Malignant, BI-RADS categories). 3. Set Up the Environment Prepare your machine learning or deep learning environment (e.g., PyTorch, TensorFlow, or Scikit-learn). Recommended preprocessing steps include: Image resizing Normalization Optional data augmentation 4. Model Development Use the selected subdataset to train binary or multiclass classification models. You may employ standard CNN architectures (e.g., ResNet, EfficientNet, VGG) or replicate model structures referenced in related literature. Ensure appropriate train/validation/test splitting for reliable evaluation. 5. Performance Evaluation Evaluate model performance using widely accepted metrics: Accuracy Precision Recall / Sensitivity F1-score AUC-ROC 6. Explainability and Visualization Enhance interpretability by applying Explainable AI (XAI) methods such as: Grad-CAM LIME SHAP These techniques help visualize decision regions and provide clinical insights. 7. Reproducibility To ensure reproducibility, document the following: Model hyperparameters Data split ratios Random seed values Library/framework versions
Institutions
- Afyon Kocatepe Universitesi