Jason Daehn Balkman MD, Presenter: Nothing to Disclose

Open source computer vision tools may be implemented in a cloud-based web application for the analysis of radiologic images. A mammographic breast density quantification tool was developed and made publicly available as a zero footprint web application. This architecture may potentially be extended to other areas of radiology, encouraging more collaborative, transparent, and standardized approaches to image processing.

Radiologic image processing is frequently performed using proprietary tools on local servers. This may limit institutional access to specialized software and lead to varied image analytics between facilities. A cloud-based platform for developing, testing, and utilizing image analysis tools could help standardize these activities and improve access to radiology software. This work focuses on breast density quantification as a potential application for such technology.

Amazon Web Services were used to launch a configurable cloud server. Open source computer vision software tools, including Python-based OpenCV, Scikit-image, Mahotas, and Scipy/Numpy were installed on the remote cloud server. Breast density analysis algorithms were implemented using these tools, capable of handling DICOM, TIFF, and JPEG image formats. Software code was uploaded to a collaborative open source repository, GitHub (github.com/jbalkman/qadense) for  reference. JavaScript and HTML scripts were used to create a front-end public website, QADense.com (Quantitative Analysis of Breast Density, pronounced “KAY-dense”). The website was connected to the configured cloud server using an NGINX webserver and Flask development environment.

The public website was freely accessible through a modern web browser such as Chrome. Mammograms could be uploaded to the website using a simple drag and drop of image files onto the website. A typical 14 MB mammogram uploaded in less than five seconds. Mammograms were analyzed by clicking a website button, with processing times under 10 seconds and all cloud server content deleted after analysis. Quantitative data, including both area and volumetric-based calculations were presented in table format, along with a visual representation of processed images.

Balkman, J, Cloud-based Implementation of an Open Source Breast Density Analysis Tool.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14003484.html