Datasets for AI Breast Cancer Detection with MWI
Explore comprehensive resources and studies on microwave imaging datasets
Highlights
- University of Manitoba Breast Microwave Imaging Dataset (UM-BMID): A primary dataset with S-parameter measurements from experimental scans.
- SAFE Microwave Breast Cancer System Dataset: Contains both benign and malignant sample data with detailed performance metrics.
- Research and Simulation Datasets: Various studies provide simulated data and numerical models using FDTD and other methods to supplement training resources.
Comprehensive Dataset Overview
When working on AI-driven detection of breast cancer using Microwave Imaging (MWI) technology, several datasets and resources can guide and enhance your research. These datasets not only provide raw measurements and imaging data but also sometimes include supplementary scripts and simulation details that are crucial for pre-processing and model training.
University of Manitoba Breast Microwave Imaging Dataset (UM-BMID)
Dataset Description and Features
The UM-BMID dataset is an open-access resource available through IEEE DataPort. It comprises S-parameter measurements taken from experimental scans using MRI-derived breast models as phantoms. This dataset offers over 1250 scans capturing a diverse range of breast densities and sizes, which makes it a versatile tool for training AI models on microwave imaging data.
The availability of this dataset and associated scripts on GitHub can facilitate your experimentation and algorithm development. It is widely referenced in studies focusing on the validation of microwave imaging systems and offers robust data for training classification models to differentiate between benign and malignant tissues.
Access the UM-BMID dataset here: UM-BMID - IEEE DataPort
SAFE Microwave Breast Cancer System Dataset
Dataset Description and Performance Metrics
The SAFE Microwave Breast Cancer System Dataset is another significant source of data for training AI models in the context of MWI. This dataset collects measurements from 113 breast samples, categorizing them into 70 benign and 43 malignant cases. This is particularly useful for studies aimed at establishing model performance based on sensitivity, specificity, and overall accuracy. For instance, the referenced study highlights a sensitivity of 79%, a specificity of 77%, and an overall accuracy of 78%.
This dataset is helpful not only for training but also for benchmarking detection algorithms in a clinical research framework. It is accessible through ScienceDirect, where further methodological details and performance criteria are thoroughly discussed.
Explore the SAFE dataset and its study here: SAFE Microwave Breast Imaging - ScienceDirect
Additional Simulation and Research Datasets
Supplementary Sources and Research Articles
In addition to the primary datasets, several research studies have established simulation methods that generate numerical data for MWI. These studies often employ Finite-Difference Time-Domain (FDTD) simulations to create electromagnetic models based on MRI-derived breast tissues. Such simulated datasets are particularly valuable when real-world data is scarce or when you wish to enhance the volume and diversity of your training set.
A notable resource is the GitHub repository "humza-sami/Machine-Learning-Based-Approaches-For-Breast-Cancer-Detection-in-Microwave-Imaging." This repository not only provides an experimental dataset but also includes detailed implementations of machine learning algorithms tailored for MWI datasets. Additionally, several academic papers and conference proceedings have mentioned the use of such synthetic data along with experimental scans to improve the robustness of classification models.
Refer to the GitHub repository for simulation datasets and approaches: GitHub Repository - Machine Learning for MWI
Organizing the Available Datasets
Summary Table of Datasets
The following table provides a concise overview of the main datasets and resources that are available for training AI models in breast cancer detection using MWI technology:
| Dataset/Resource | Description | Access Link |
|---|---|---|
| UM-BMID | S-parameter measurements from MRI-derived breast phantoms; over 1250 experimental scans. | IEEE DataPort |
| SAFE Microwave Breast Cancer Dataset | Dataset with 113 samples (70 benign, 43 malignant) with performance metrics for MWI imaging. | ScienceDirect Article |
| GitHub Repository for MWI Detection | Contains simulated data and machine learning implementations for breast cancer detection using MWI. | GitHub Repository |
| Additional Research Publications | Includes studies using FDTD simulations and clinical trial data offering supplementary data for MWI. | MDPI Study |
Accessing and Utilizing These Datasets
It is essential when accessing these datasets to note the following:
Permissions and Terms
Several datasets are open-access; however, some might have restrictions or require registration, specific usage citations, or collaboration requests. Always review the dataset's terms of use and ensure that any experiments or publications citing the data adhere to the associated guidelines.
Integration for AI Training
Incorporating data from multiple sources can strengthen your model’s performance by providing a broader range of cases. Whether through experimental scanning data from UM-BMID or simulation data via FDTD methodologies, utilizing a mixture of data types can help in reducing overfitting and increasing generalization. Advanced pre-processing techniques such as normalization, noise reduction, and data augmentation are recommended to further improve training efficiency.
Simulated Versus Experimental Data
It is often beneficial to combine experimental datasets with simulated data. While real-world data is crucial for understanding natural variances and noise characteristics, simulated datasets can fill in gaps where experimental data may be insufficient. Moreover, simulations allow you to create controlled scenarios that may not be readily available in experimental studies, thus providing additional training opportunities.
References and Further Reading
To dive deeper into this subject and explore the latest developments in microwave imaging for breast cancer detection, consider reviewing these references:
- UM-BMID - IEEE DataPort
-
SAFE Microwave Breast Imaging - ScienceDirect
- GitHub - MWI Detection Repository
- MDPI Study on MWI
-
Clinical Investigation of MWI
Recommended Related Queries
Last updated March 13, 2025