Menu Content/Inhalt
Main arrow Research arrow Trabajos arrow Microcalcification cluster diagnosis in digitized mammograms
Microcalcification cluster diagnosis in digitized mammograms PDF
Written by Ramón Gallardo Caballero   
Article Index
Microcalcification cluster diagnosis in digitized mammograms
Our Proposal
Methodology
System implementation
Results
Future developments

Implementation

As can be deduced from the previous paragraphs, the proposed system is very complete and somewhat complex. In this paragraph we describe in deep the elements which constitute the system and its characteristics.

DDSM server

The database provided by Island Vision Group is accessible by FTP, so we dumped it to a dedicated server with a storing capacity of 290 gigabytes where 240 gigabytes are used exclusively to store the image and configuration files which constitutes the database.

PostgreSQL database

As previously mentioned, DDSM provides case information in file form. As this format isn’t adequate to work in an intensive way with this data we have decided to insert this information in a SQL database server which runs in one of our Linux servers.

At the same time this database engine let us store result information for different project phases, letting us to centralize all needed data, both input as output; and facilitating statistics procurement task.

Computational infrastructure

Some project phases require high computational resources and volatile memory requirements. Fortunately our research group owns various multiprocessor servers with important memory resources which have allowed us to carry out the project; obviously typical hardware infrastructure is insufficient to deal with a project like the proposed here.

We can point out an Opteron bi-processor server with 16 gigabytes of RAM, a second 64-bit server Athlon64 mono-processor with 4 gigabytes of RAM and various 32-bit bi-processor servers which holds the database engine, the distributed file system and backup.

A special mention is due to the most important computation resource of the group, a Beowulf type cluster with 45 AthlonXP-2000 nodes each one equipped with 512 megabytes of RAM. This is the resource where all massive calculations are carried out and let us reduce dramatically simulation times from 40 to 45 times.

The management interface: MAMOPROT

A reliable image work requires developing a tool which let to visualize these images and to make different operations with them like marking or correction. The tool developed by our group is called Mamoprot and let us to solve many different issues arisen in various development stages of the project.

Mamoprot is a modular application written in C++ and developed against the widget library QT which runs under Linux environment. It’s completely integrated with the relational database server which manages the project. This tool lets to carry out many different operations, so besides basic visualization operations (contrast adjustment, brightness, basic filtering and enhancement) working with overlays let us to carry out many interfacing operations. Specifically Mamoprot is the tool which let us visualize and mark prototypes in overlays, also let us to diagnose the visualized overlay with a given configuration, visualize ICA decomposition of the overlay and rebuild the image using only some components. The first module of Mamoprot was used by experts to insert training prototypes in database.

Mamoprot also is designed to deal with mammograms letting us to visualize them with overlapped expert marks and stored regions of interest. Also let us to carry out a diagnostic of the whole mammogram viewing pixel based classification and the generated ROIs with that system configuration. Finally this module allows us to create new overlays to add new prototypes for interesting regions.

And finally, Mamoprot allows to visually inserting new diagnostic configurations selecting system characteristics to be studied. In the following figure we can see a Mamoprot snapshot for the mammogram module; particularly we show the diagnostic results for a complete mammogram. Discontinued blue rectangles mark benign auxiliary ROIs previously added to the database, red discontinuous rectangle marks the original DDSM malignant overlay, green solid line rectangle accounts for true positive finding and the blue solid border rectangle is a false positive generated by the system. You can click on the image below to get a detailed version in a new window.

Mamoprot

 



Last Updated ( jueves, 25 noviembre 2010 )