Usage: This GAN is used as part of the medigan library. This GANs metadata is therefore stored in and retrieved from medigan-model's config file. medigan is an open-source Python library on Github that allows developers and researchers to easily add synthetic imaging data into their model training pipelines. medigan is documented here and can be used via pip install: pip install medigan To run this model in medigan use the following commands. # import medigan and initialize generators from medigan import Generators generators = Generators() # Generate 10 images generators.generate(model_id="00001_DCGAN_MMG_MASS_ROI",num_samples=10) Description: A deep convolutional generative adversarial network (DCGAN) that generates regions of interest (ROI) of mammograms containing benign and/or malignant masses. Pixel dimensions are 128x128. The DCGAN was trained on ROIs from the Optimam dataset (Halling-Brown et al, 2014). The uploaded ZIP file contains the files malign_mass_gen (model weights), and __init__.py (image generation method and pytorch GAN model architecture). Metadata: The source of the metadata displayed below is the global.json file in the medigan-models Github repository. { "00002_DCGAN_MMG_MASS_ROI": { "execution": { "package_name": "MALIGN_DCGAN", "package_link": "https://zenodo.org/record/5189243/files/MALIGN_DCGAN.zip?download=1", "model_name": "malign_mass_gen", "extension": "", "image_size": [ 128, 128 ], "dependencies": [ "numpy", "torch", "opencv-contrib-python-headless" ], "generate_method": { "name": "generate", "args": { "base": [ "model_file", "num_samples", "output_path", "save_images" ], "custom": {} } } }, "selection": { "performance": { "turing_test": { "number_radiologists": 2, "AUC": [ 0.56, 0.45 ], "accuracy": [ 0.48, 0.61 ], "years_experience": [ 7, 25 ] }, "downstream_task": { "CLF": { "trained_on_real_and_fake": { "fraction_real_data": 0.067, "f1": 0.89 }, "trained_on_real": { "fraction_real_data": 0.067, "f1": 0.89 } } } }, "use_cases": [ "classification" ], "organ": [ "breast", "breasts", "chest" ], "modality": [ "MMG", "Mammography", "Mammogram", "full-field digital", "full-field digital MMG", "full-field MMG", "full-field Mammography", "digital Mammography", "digital MMG", "x-ray mammography" ], "vendors": [ "Hologic Inc" ], "centres": [], "function": [ "noise to image", "image generation", "unconditional generation", "data augmentation" ], "condition": [], "dataset": [ "Optimam" ], "augmentations": [], "generates": [ "mass", "masses", "breast masses", "mass rois", "mass ROIs", "mass images", "breast mass ROIs" ], "height": 128, "width": 128, "depth": null, "type": "DCGAN", "dataset_type": "public", "license": "MIT", "privacy_preservation": null, "year": "2019", "tags": [ "Turing Test", "Visual Turing Test", "Mammogram", "Mammography", "Digital Mammography", "Full field Mammography", "Full-field Mammography", "128 x 128", "128x128", "MammoGANs", "Nodule", "Nodules", "Breast mass" ] }, "description": { "title": "DCGAN Model for Mammogram Mass Region of Interest Generation (Trained on OPTIMAM)", "provided_date": null, "trained_date": null, "provided_after_epoch": null, "version": null, "publication": null, "doi": [ "10.5281/zenodo.5188557", "10.1117/12.2543506", "10.1117/12.2560473" ], "comment": "A deep convolutional generative adversarial network (DCGAN) that generates regions of interest (ROI) of mammograms containing benign and/or malignant masses. Pixel dimensions are 128x128. The DCGAN was trained on ROIs from the Optimam dataset (Halling-Brown et al, 2014). The uploaded ZIP file contains the files malign_mass_gen (model weights), and __init__.py (image generation method and pytorch GAN model architecture). Kernel size=6 used in DCGAN discriminator." } } }