Skip to content

Latest commit

 

History

History
140 lines (87 loc) · 11.5 KB

README.md

File metadata and controls

140 lines (87 loc) · 11.5 KB
Raw

ddsm — Digital Database for Digital Mammography Software

Dockerisation

There is a docker version of this repo that will transform all the ddsm images to png.

Preparation - obtain the data

Use the included ftp_all.sh command (or similar) to do the download. The results will form a tree starting with the USF hostname. Use this directory as a parameter when invoking the docker conversions image instead of <USF_TREE>

EASY WAY - Invoke the docker image from dockerhub

docker run -v <USF_TREE>:/ddsm_data topiaruss/ddsm

Alongside the LJPEG images in the tree the png equivalent will be created.

Note that the PNG files are roughly double the size of the originals, so your data will grow to 3x the size once processing is complete.
At this point you have the option to delete the originals:

find <USF_TREE> -name "*LJPEG" -delete

DEVELOPER'S WAY - building the dockerfile locally

cd ddsm
docker build -t ddsm . ; docker run -v <USF_TREE> ddsm

Introduction

This file describes how to install and use software to:

  • download DDSM mammograms and convert them into a usable format;
  • obtain the annotations (e.g., tumour segmentations) made by the DDSM radiologists and the associated metadata (i.e., malignancy type etc.)

Please note that while this GitHub repository was published in June 2016, the included software, instructions below, and the PDF manual are now pretty old (as are the digitised mammograms!), and absolutely no support is offered. However, this software (at its original home) was still being downloaded around 100 times per month as of June 2016. Please also note that this GitHub repository is not affiliated with the DDSM project, and questions about the database should be directed to the DDSM project’s maintainers at the University of South Florida.

While it may appear from the following description that the process of obtaining mammograms and their annotations is a highly manual process, the steps involved are presented to allow users to understand what is atually happening. The software is intended to be called by your own programs or scripts, as your workflow necessitates.

Copyright, Licensing, Warranty, and Fitness for Use

This repository contains software that is in the public domain, is publicly available, or has been specially adapted or developed to be used within the Windows and Cygwin environment. You are granted a non-exclusive license to use or modify the software for research purposes only. You are encouraged to submit pull requests if you can improve this software for the research community.

The accompanying software must be used for research purposes only. It has not been validated for use in clinical or similar environments. There is no warranty for this software, which is supplied “as is”, and no claims are made about fitness for any purpose. Portions of the software have, however, been used by many computer-aided mammography researchers for many years.

Dependencies

The software has been tested on:

  • Windows XP Professional with Service Pack 2.
  • Linux (x86, 64-bit only).

Later versions will probably work, but are untested.

If you are using Windows, you will also need the Cygwin UNIX environment (version 1.5.25-15 or later; installation instructions are available in [DDSM User Manual.pdf](DDSM User Manual.pdf)).

You will also need:

  • ImageMagick (version 6.4.0.6-1 or later).
  • Ruby (version 1.8.7-p72-2 or later).
  • MATLAB (version 7.6.0 (R2008a) or later).
  • An FTP client (one is built in to Windows Explorer, though other options are available).

The [DDSM User Manual](DDSM User Manual.pdf) contains installation instructions for ImageMagick and Ruby (for Windows), but these may be very out of date now.

Installation Instructions

The following instructions assume you have installed the dependencies above.

  1. Clone this repository (or download it as a zip file and unzip it). Linux users can place their local copy wherever is convenient. Windows users should place their local copy in the Cygwin home directory (i.e., C:\cygwin\home\<Your User Name>).
  2. User the terminal (i.e., the Cygwin window for Windows users) to change directory to the dddm-software directory using the command cd ddsm-software.
  3. Verify that the correct files are present. Run ls and ensure that the following files are listed: ddsmraw2pnm.exe, get_ddsm_groundtruth.m, get-ddsm-mammo, info-file.txt, jpeg.exe, and jpeg.
  4. Verify you can run the program that converts the mammograms. Run ./jpeg (Linux users) or ./jpeg.exe (Windows users). You should see some usage instructions printed to the screen. (You will never need to run this command directly, however, so you can ignore them.) If the program does not run, you may need to set the file to be executable (e.g., using the chmod command).
  5. Verify that Ruby was installed correctly. Run ruby -e “p ‘Hello world!’” (don’t miss the single quotes!). You should see “Hello world!” printed to the screen.

The DDSM radiologist annotations and metadata (which includes all information provided about the annotations in the .OVERLAY files, such as the type of abnormality and its subtlety) are made available via the get_ddsm_groundtruth.m MATLAB function. Install this software as follows:

  1. Copy or move the file get_ddsm_groundtruth.m from the dddm-software directory to a directory in which you keep your MATLAB software (create such a directory if one does not exist).
  2. Start MATLAB.
  3. Add the directory into which you placed the get_ddsm_groundtruth.m file in step 1 to your MATLAB path (see What Is the MATLAB Search Path?).
  4. From the MATLAB command prompt, check that MATLAB can find the get_ddsm_groundtruth.m file by running help get_ddsm_groundtruth. MATLAB should print the documentation for the get_ddsm_groundtruth function.

How to Obtain and Convert a DDSM Mammogram

The software provided makes it simple to get a DDSM mammogram in a usable file format. The following example shows how we can get the mammogram A_1141_1.LEFT_MLO in PNG format. PNG format files can be read by a wide range of software, including MATLAB, Photoshop, and Windows itself. It is a lossless file format and offers good compression ratios. (Note, however, that each DDSM mammogram is about 40MB.)

To download and convert the mammogram A_1141_1.LEFT_MLO into PNG format:

  1. If you closed the terminal or Cygwin program, restart it and change to the dddm-software directory.
  2. Run ./get-ddsm-mammo A_1141_1.LEFT_MLO. The mammogram will be downloaded and the converted file will be placed in the ddsm-software directory. The full (Unix) path to the file will be printed to the screen. You should probably move the resulting files to a more convenient directory.
  3. View the downloaded and converted file. For example in Windows, double-click on the file.

The get-ddsm-mammo step obtains and converts the mammogram named A_1141_1.LEFT_MLO. The program connects to the DDSM’s FTP server, downloads the corresponding “lossless” JPEG file, converts that file to a raw binary format, converts that file to a simple human-readable file format called PNM, converts that file to the desired PNG format, and finally deletes the “lossless” JPEG and all intermediate files. Because the DDSM files are large and conversion is processor intensive, it may take a nontrivial amount of time for the file to be downloaded and converted. Exactly how long depends on the speed of your Internet connection and how powerful your computer’s CPU is. When this software was originally written (c. 2006), it took about 3 minutes to download and convert a mammogram using a domestic 10Mbps cable modem and a 2006-vintage 2GHz Intel Core 2 Duo CPU. This will hopefully be appreciably faster for 2016-vintage Internet connections and CPUs.

If the get-ddsm-mammo program is interrupted, intermediate files may be left in the ddsm-software directory. Such files will have the suffix .1 or .pnm, and should be deleted as they tend to be relatively large (80MB+).

Note that, as of c. 2006, the DDSM’s FTP server had a policy of allowing no more than 10 users at a time. If the get-ddsm-mammo program fails, the limit on the number of users is a possible reason. In the first instance, simply wait a few minutes and try again. (While testing this software, the DDSM’s FTP server went offline for several hours, so be aware that this may be a “weak link” in your workflow.)

How to Obtain DDSM Radiologist Annotations and Metadata

For this example, we will obtain the annotations and metadata for the file A_1580_1.LEFT_MLO. This mammogram was chosen because it is an example of a non-trivial mammogram: it has multiple abnormalities (boundaries) and one of those has a core annotation. You can obtain ground truth (annotations and metadata) for this file as follows:

  1. Using your FTP client (such as Windows Explorer), connect to the DDSM FTP server at figment.csee.usf.edu. If prompted, use the username “anonymous” and an empty password.
  2. Navigate to the directory /pub/DDSM/cases/cancers/cancer_10/case1580/.
  3. Copy the remote file A_1580_1.LEFT_MLO.OVERLAY to a convenient directory on your local computer. Let us call this directory <overlay-directory>.
  4. Disconnect from the DDSM’s FTP server so that you aren’t preventing other users from accessing it.
  5. At MATLAB’s command prompt, run cd <overlay-directory>.
  6. At MATLAB’s command prompt, run ground_truth = get_ddsm_groundtruth(‘A_1580_1.LEFT_MLO.OVERLAY’);. The rest of this tutorial is provided by the get_ddsm_groundtruth function’s documentation (read from “Example usage”). To see the documentation, run help get_ddsm_groundtruth.

Note that to get a binary mask image showing the region annotated by the radiologist, you will need to know the image dimensions. You can obtain this information in one of two ways:

  • Use the DDSM’s website and view the thumbnails pages. These list the contents of the corresponding case’s ICS file, which gives the number of lines (i.e., rows) and pixels per line (i.e., columns) for each mammogram in the case.
  • Obtain the corresponding mammogram (using the get-ddsm-mammo program) and use MATLAB’s size function to get the number of rows and columns in the mammogram.

Parsing the .OVERLAY file is very quick, but creating a binary mask matrix using the anonymous function(s) returned by get_ddsm_groundtruth.m can take a couple of seconds.

Saving the Annotations

The get_ddsm_groundtruth.m code returns a cell array of structs, which can be saved to disk using MATLAB’s save command. Such files will require very little disk space relative to the full binary mask matrices (which are the same size as their corresponding mammograms).

What Can Go Wrong

Apart from the data and this software now being very old, the weakest link in the workflow is the DDSM’s FTP server, which may permit a limited number of users to access the server at any given time.

The get-ddsm-mammo program should be quite robust, with the exception of its reliance on the DDSM’s FTP server. If you experience problems using it, please try connecting to the DDSM’s FTP server using your FTP client, to see if this is the cause of the problem.