eDiamond and Mammogrid to develop grid infrastructure for standardised annotated mammograms

Lyon 16 January 2003At the First European HealthGrid Conference, Jérôme Declerck, chief scientist at Mirada Solutions, presented the British eDiamond and the European Mammogrid projects, both dealing with the screening and detection of breast cancer. These two-year initiatives were showcased as potential prototypes for supranational databases of medical images since both project consortia are trying to build grid-enabled federated databases of annotated mammograms and testing validation of the need for standardisation of fixed parameters to detect breast tumours using Mirada's Standard Mammogram Form (SMF) in eDiamond.

Advertisement

Mammogrid as well as eDiamond aim at demonstrating the impact of image standardisation for medical image databases for mammography, cancer in general, degenerative brain diseases and heart dysfunctionalities. As such, the "find-one-like-it" data mining method will be applied to generate examples of annotated masses, as the speaker explained.

The British eDiamond consortium comprises the University of Oxford, the University College of London, King's College of London, the University of Edinburgh, St. George's Hospital of the University of London, Oxford Radcliffe Hospitals, the Scottish Breast Screening Programme, IBM, and Mirada Solutions. eDiamond stands for Digital Mammography National Database. The EC-funded Mammogrid project consortium has eight partners, involving the Universities of Bristol, Oxford, Cambridge, Sassari, Pisa, and Udine; CERN; and once again Mirada Solutions.

Both projects are complementing and reinforcing each other. Whereas eDiamond concentrates on teaching, tele-diagnosis and algorithm development for data mining to influence the future of breast screening programmes by encouraging them to embrace and pioneer the SMF technology; Mammogrid is focusing on quality control for breast cancer screening, epidemiology of breast cancer from a European perspective, and the creation of CADe algorithms.

The electronic health records as we know them display a large amount of text but since 1990, there is a growing and urging demand for medical images to better understand the morphology and physiology of breast cancer in particular and disease in general, as Mr. Declerck outlined.

The speaker showed some statistics that highlighted the need to undertake action and fight breast cancer. One in eight women are bound to develop the disease during their lives, and one in 28 will die of it. In the European Union and the United States, 348.000 women are diagnosed with it and it kills 115.000 patients each year. There were 900.000 new cases worldwide in 1997 and in the European Union, breast cancer accounts for 19 percent of all cancer victims and 24 percent of cancers detected. These statistics are currently being replicated in Asia. It goes without saying that the sooner the diagnosis is made, the better the prognosis will be.

The British Screening Programme started in 1987. Every three years, all women aged between 50 and 64 are invited to undergo screening. In this way, some 1,5 million women in England are screened annually in 92 centres, totalling 230 radiologists. The average radiologist sees about 5000 cases per year. Adverse signs are detected in about 30 mammograms per 1000. Although a double reading is required by a radiologist as well as a senior radiographer, still 20 percent of the cancers are missed.

Mr. Declerck stressed that regular breast screening programmes lead to a flood of data. Currently one view taken of each breast amounts to 3.000.000 images per year, increasing to two views per breast over the next two to five years, which makes lossless compression of these images indispensable. In addition, a host of variables seriously affects the appearance of a mammogram. This calls for standardisation because the dense breast tissue is extremely difficult to detect. Irrelevant variations can swamp those that are clinically significant. As a result, images submitted to a database can tell more on the radiographer's competence and the equipment's age than on the incidence of cancer.

Mirada's Standard Mammogram Form (SMF) technology as a common denominator for image comparison and deployed over a grid network gives rise to a new product for differential diagnosis, according to the speaker. The "find-one-like-it" method will improve diagnostic confidence and can be used to monitor diagnostic quality and teaching. The challenges for both project consortia are to create large federated databases containing images, ontologies, and metadata including image features and formation parameters, and clinical and demographic data; to generate effective data mining; and enable complex queries with a range of executables.

The issues that need to be solved consist in handling the legal restrictions on access to data that resides in hospitals and is firewall-protected; combining a variety of databases; securing file transfer of large images; and developing application programming interfaces (API) for black box third party applications.

There are several options to host the data in a grid-enabled database. First of all, patient data can be stored centrally while the images remain at the hospitals. Second, portions of the central database could be copied to each of the hospitals. Third, it is possible to create a logical rather than a physical database in the central server hosting a federated view of the data at the hospitals. Whenever there is a query, the database management structure automatically copies the query out to the databases at the hospitals and collates the results afterwards. Fourth, a cross-federated database could be installed in which every hospital has a federated view of all the patient data in the system.

Mr. Declerck concluded that a number of open questions remains to be solved by eDiamond and Mammogrid. There is the question of finding the right way to federate the database and to guarantee the security, required by the law, without reducing the database performance. There is the issue of the hospital firewalls that need to be overcome. There is the problem of creating an appropriate metadata format which allows to retrieve the data in a fast manner. And there is also the need to ensure the system's Quality of Service and fault-tolerance.

Ambitious grid initiatives as eDiamond and Mammogrid may be, they are not the final solution to a perfect health care informatics infrastructure nor to a fully computer-aided detection of breast cancer. If both projects lead to a completely functional and scaleable prototype with user involvement from the start and with novel medical image analysis research deliverables, this would already be a terrific result. More details about the eDiamond project can be found in the VMW November 2002 article Oxford University, IBM, and British Government build eDiamond Grid for breast cancer screening and diagnosis.


Leslie Versweyveld

[Medical IT News][Calendar][Virtual Medical Worlds Community][News on Advanced IT]