DCGAN Model for Mammogram Calcification Region of Interest Generation (Trained on INbreast)

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_CALC_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 calcifications. Pixel dimensions are 128x128. The DCGAN was trained on ROIs from the INbreast dataset (Moreira et al, 2012).

The uploaded ZIP file contains the files dcgan.pt (model weights), __init__.py (image generation method and utils), a README.md, and the GAN model architecture (in pytorch) below the /src folder.

Metadata:

The source of the metadata displayed below is the global.json file in the medigan-models Github repository. 

   
{
  "00001_DCGAN_MMG_CALC_ROI": {
    "execution": {
      "package_name": "DCGAN",
      "package_link": "https://zenodo.org/record/5526998/files/DCGAN.zip?download=1",
      "model_name": "DCGAN",
      "extension": ".pt",
      "image_size": [
        128,
        128
      ],
      "dependencies": [
        "numpy",
        "Path",
        "torch",
        "opencv-contrib-python-headless"
      ],
      "generate_method": {
        "name": "generate",
        "args": {
          "base": [
            "model_file",
            "num_samples",
            "output_path",
            "save_images"
          ],
          "custom": {
            "image_size": 128
          }
        }
      }
    },
    "selection": {
      "performance": {
        "SSIM": null,
        "MSE": null,
        "NSME": null,
        "PSNR": null,
        "IS": null,
        "FID": null,
        "turing_test": null,
        "downstream_task": {
          "CLF": {
            "trained_on_fake": {
              "accuracy": null,
              "precision": null,
              "recall": null,
              "f1": null,
              "specificity": null,
              "AUROC": null,
              "AUPRC": null
            },
            "trained_on_real_and_fake": {},
            "trained_on_real": {}
          },
          "SEG": {
            "trained_on_fake": {
              "dice": null,
              "jaccard": null,
              "accuracy": null,
              "precision": null,
              "recall": null,
              "f1": null
            },
            "trained_on_real_and_fake": {},
            "trained_on_real": {}
          }
        }
      },
      "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": [],
      "centres": [],
      "function": [
        "noise to image",
        "image generation",
        "unconditional generation",
        "data augmentation"
      ],
      "condition": [],
      "dataset": [
        "INbreast"
      ],
      "augmentations": [
        "crop and resize",
        "horizontal flip",
        "vertical flip"
      ],
      "generates": [
        "calcification",
        "calcifications",
        "calcification roi",
        "calcification ROI",
        "calcification images",
        "calcification region of interest"
      ],
      "height": 128,
      "width": 128,
      "depth": null,
      "type": "DCGAN",
      "license": "MIT",
      "dataset_type": "public",
      "privacy_preservation": null,
      "tags": [
        "Mammogram",
        "Mammography",
        "Digital Mammography",
        "Full field Mammography",
        "Full-field Mammography",
        "128x128",
        "128 x 128",
        "MammoGANs",
        "Microcalcification",
        "Microcalcifications"
      ],
      "year": "2021"
    },
    "description": {
      "title": "DCGAN Model for Mammogram Calcification Region of Interest Generation (Trained on INbreast)",
      "provided_date": "12th May 2021",
      "trained_date": "May 2021",
      "provided_after_epoch": 300,
      "version": "0.0.1",
      "publication": null,
      "doi": [
        "10.5281/zenodo.5187714"
      ],
      "comment": "A deep convolutional generative adversarial network (DCGAN) that generates regions of interest (ROI) of mammograms containing benign and/or malignant calcifications. Pixel dimensions are 128x128. The DCGAN was trained on ROIs from the INbreast dataset (Moreira et al, 2012). The uploaded ZIP file contains the files dcgan.pt (model weights), __init__.py (image generation method and utils), a README.md, and the GAN model architecture (in pytorch) below the /src folder. Kernel size=6 used in DCGAN discriminator."
    }
  }
}