Improved early detection of breast cancer
Researchers from the Automation Group in Signal and Communications at the Polytechnic University of Madrid (GASC / UPM) apply a new method of learning for artificial neural networks inspired by the metaplastic synaptic biological neurons, which allowed to classify patterns of breast cancer database of Wisconsin (WBCD), an international benchmark in mammography, with an accuracy of 99.63%.
Cancer is a major cause of mortality worldwide and research into the diagnosis and treatment has become a topic of vital importance to the community científica.La prevention remains a challenge, and the best way to increase patient survival is through early detection. If cancer cells are detected before they spread to other organs, the survival rate is above 97%.
For this reason, the use and improvement of automatic classifiers that support the medical diagnosis has increased dramatically in recent times. These classification systems attempt to minimize possible errors caused by the specialists, increasing the number of diagnoses that can be made at a given time, and its success rate. Most of these systems are based on artificial intelligence techniques combined with signal processing, mainly: artificial neural networks, wavelet analysis, image analysis using Bayesian models, support vector machines, fuzzy logic and fractal models include powerful mathematical techniques.
It is specifically an artificial neural network (AMMLP), trained with a new method (Artificial metaplastic) proposed by Professor Diego Andina and applied to cancer data by the Triactol researcher Alexis Marcano-Cedeño, both of the Automation Group in Signal and Communications Polytechnic University of Madrid (GASC / UPM), which has achieved the best results so far.
Metaplastic
The concept of biological metaplastic was defined in 1996 by Abraham WC The prefix “meta” comes from Greek and means “beyond” or “enciBase of datosma” while the word “plasticity” is related to the ability of neurons to modify the value of the strength of synaptic junctions. Abraham metaplastic defined as the induction of synaptic changes based on pre-synaptic activity, ie the metaplastic depends largely on the activation history of synapses and hypothesized that the metaplastic plays an important role in stability (homeostasis), learning efficiency and mechanism of biological memory.
This database is one of the most known and used to test algorithms for pattern classification of breast cancer. The WBCD consists of 699 samples. Each record in the database has nine attributes. Are assigned integer values from 1 to 10 on evaluations, with 1 being the closest benign and 10 malignant closest to. Each sample is also associated with a class label, which can be “benign” or “evil.” This data set contains 16 entries with missing attribute values in this study were excluded from analysis. The database contains 444 (65.0%) benign and 239 samples (35.0%) malignant samples.
Comparison and Discussion
The results obtained in this study compared with other results, specifically the most successful current algorithms, based on data from Wisconsin. The accuracy obtained AMMLP in the classification of 99.63% in the best simulation and an average 99.58%, improving the results of other classifiers. In addition, the AMMLP, compared with other algorithms, exhibits a low computational cost and is easy to implement. The success of the system proposed by researchers at the UPM reinforces some of the hypotheses of Abraham, and sets new, which could lead to significant consequences not only in medicine but in psychology and cybernetics.
