90% Precision in Detecting Breast Cancer
By: Johanna Ruiz Bolaños, Unimedios
With greater precision than in conventional methods for detecting cancerous masses, biomedical engineers at the Universidad Nacional (National University) launched their Support System for Breast Cancer Diagnosis, a disease that affects 7,000 Colombian women each year (according to the 2010 National Survey on Demography and Health / Encuesta Nacional de Demografía y Salud 2010) and which can be cured as long as it is discovered in time.
This system, as explained by biomedical engineer Fabián Narváez, “can detect masses in 80.3% of cases and determine the classification of pathology or severity, in other words, whether it is benign or malignant, with 85.3% precision”.
He clarifies that the system is based on single projection mammography, or in other words that screening is carried out on women who do not show any symptoms, but have perceived some type of anomaly when they examine themselves. “X–rays are used to create two projections of each breast in order to have two perspectives”.
With those mammographs, radiologists can see each breast and compare them. “The idea is to compare the images and make an inspection. The supposition is that, given their nature, there must be symmetry and a similar distribution of tissues”, according to Narváez.
With this approach, the experts seek zones that have been affected or show indications of masses. That is the idea behind the system implemented by the UN: superimposing the images provides conclusive results.
Less margin for error
Masses are a constant headache for experts, as existing methods are not totally effective because while they easily detect microcalcifications (the more advanced tumors), they do not detect masses that are frequently confused with other types of afflictions. Also, this is not the only problem. Breast implants, which are very popular, can sometimes affect the results.
But this new platform will make it possible for physicians, whether specialists or not, to have a much more reliable technique to detect an illness that if not caught in time can be fatal. Narváez explains that the objective of the Support System for Breast Cancer Diagnosis is to provide an effective tool in even the most remote zones of the country, where it is practically impossible to gain access to an expert.
Differences between the CADS and the Support System
The new method was created under the Bi–rads Protocol, use by radiologists to identify and compare mammographic masses. According to the biomedical engineer, depending on the form and size of the mass, it is possible to determine whether it is benign or malignant and suggest timely treatment.
“This led us to analyze the general functioning of the systems. There are currently various platforms called CADS, or Computer–Assisted Diagnosis, which serve as automatic tools to detect or segment regions of interest. They are guides to enable radiologists to easily identify if there is an abnormality in the breasts”.
The problem with the CADS is that, because they use mathematical algorithms, there is the risk that detection will not be exact. “Depending on the volume of cases, a bias can be created which leads to various errors”, according to Narváez.
This happens because information is used from one single projection, whereas the support platform makes it possible to superimpose two images at the same time. Also, the CADS system does not make annotations, because it merely locates the region where the mass can be found. “Our system shows the characteristics of the masses, their size, form and location. This is something completely new”, stresses Fabián Narváez.
According to studies carried out by the Telemedicine Group at the National University (Grupo de Telemedicina de la UN), CADS have 85% sensitivity in detecting microcalcifications, but only 65% in detecting masses, a relatively low percentage which means that these tools are not widely accepted in real clinical scenarios.
“In addition to describing the region in radiological terms, the objective is to suggest a diagnosis, although ultimately it is the radiologist who makes that decision”, according to the engineer.
This is how it works
The two projections that have been made are used, one from above and one in profile. They are then superimposed, and the mass is analyzed based on the observations.
The process is used to obtain an image of crisis levels, which is then used to make one description of the form and another of the texture to determine whether or not the mass is malignant. Its contours are determined compared to the rest of the tissues and the decomposition of the image is obtained. This helps to define five characteristics: texture, contrast, porosity, form and the details of the edges, which makes the information even more valuable.
The system is based on image recovery through visual content, similar to the platform used by Google when searching for photographs. “We generate a database with cases already diagnosed and verified through biopsies”, says Narváez.
When a new case appears, the radiologist marks the region of interest and the system compares it with already–diagnosed cases. Then a set of images is found indicating similarities with small tumors. This comparison measure is called KNN.
The radiologist then uploads the images into a database on the National University Medical Faculty webpage, which subscribes to a public domain of the University of South Florida that has a record of 2,600 case involving breast cancers, with detailed graphic information along with their characteristics. All of those data make it possible to suggest a new diagnosis.