• OpenDiamond

• OpenDiamond

• 64-bit environment
• OpenDiamond
• virtualenv
• OpenCV-Python
• BLAS library and headers
• LAPACK library and headers
• freetype library and headers
• libpng library and headers
• libjpeg library and headers

1. Use make-filter.py to build the Autosegment filter.
2. Use make-filter.py to build the Gemini filter.
3. Create and populate virtualenv. Use --system-site-packages, since OpenDiamond can't be installed with pip.
4. Configure and run. dermshare.py is always needed, but dermshare-ws.py is only necessary if the mobile device link is enabled with WEBSOCKET_BASEURL.

To avoid conflicts with older versions of NumPy and Six packaged by your Linux distribution, it is best to (re-)install NumPy v1.9.1 and Six v1.9.0 with pip after installing OpenCV-Python.

fil_autosegment and fil_gemini must install dependencies into a virtualenv in /tmp the first time it runs. To avoid long delays in the first search, the virtualenv can be prepopulated by running fil_autosegment and fil_gemini without arguments on the Diamond servers. To avoid additional delays whenever a Diamond server is rebooted or /tmp is cleared, you can globally install the necessary Python libraries:

pip install -r autosegment/requirements.txt
pip install -r gemini/requirements.txt

An example scope server is provided in simplescope. The scope server shipped with OpenDiamond will not work because it doesn't integrate with Diamond web apps.

An example dataretriever is provided in dataretriever. It includes support for serving segmentation images and pathology reports in addition to the Diamond objects and scope lists. The diamond_store dataretriever shipped with OpenDiamond will probably also work, but has not been tested.

The DermShare dataretriever expects a directory tree with the following layout:

DERMSHARE_SCOPE_PATH
    <scope1>
        <name1>-image.{jpg|png|...}
        <name1>-segm.png
        <name1>-attr.yaml
        <name1>-path.html
        <name2>-image.{jpg|png|...}
        ...
    <scope2>
        ...

The segmentation image must be a PNG the same size as the image, with the pixels corresponding to the lesion marked in red and the remaining pixels transparent.

The attributes file, if present, contains only one attribute, the diagnosis code as listed in search.js:

diagnosis: bcc

The path file, if present, contains the pathology report to be shown in DermShare.