Paraconsistent Artificial Neural Network Applied in Breast Cancer Diagnosis Support
verfasst von
:
Fábio Vieira do Amaral, Jair Minoro Abe, Alexandre Jacob Sandor Cadim, Caique Zaneti Kirilo, Carlos Arruda Baltazar, Fábio Luís Pereira, Hélio Côrrea de Araújo, Henry Costa Ungaro, Lauro Henrique de Castro Tomiatti, Luiz Carlos Machi Lozano, Renan dos Santos Tampellini, Renato Hildebrando Parreira, Uanderson Celestino, Rafael Espirito Santo, Cristina Corrêa Oliveira
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden.
powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden.
powered by
Abstract
In this work, a Paraconsistent Classifier for the diagnosis of breast cancer based on the attributes of mamographic images was developed. The system uses a neural network with decision making from the final results of processing each set of attributes was created. In order to mitigate the effects of false positives diagnoses and true positives. In order to analyze the performance of the Paraconsistent Classifier, the results will be compared with the results of the following classifiers: Multi-Layer Perceptron (MLP), a dual stage classifier (ART2LDA) based on Adaptive Resonance Theory (ART) and a classifier implemented with nonlinear optimization techniques and combinatorics, associated with the classification capabilities of Radial basis Functions - (RBF-Simulated Annealing).To perform the simulations, two different databases were used. The first one, to classify calcifications, is composed of 143 samples divided into 64 benign cases and 79 malignant cases represented by form. The performances of the classifiers in discriminating benign and malignant cases are compared in terms of area under the Receiver Operating Characteristic Curve (Az). The higher the value of Az, the better the performance of the classifier.The experiments with calcifications show: Paraconsistent Classifier (Az = 0.986), MLP classifier (Az = 0.70), ART2LDA Classifier (Az = 0.696) and RBF Classifier - Simulated Annealing (Az = 0.94). For experiments with mammographic masses and tumors show: Set 1, Paraconsistent Classifier (Az = 0.939), MLP classifier (Az = 0.994), ART2LDA Classifier (Az = 0.901) and RBF Classifier - Simulated Annealing (Az = 0.912). Set 2, Paraconsistent Classifier (Az = 0.935), MLP classifier (Az = 0.994), ART2LDA Classifier (Az = 0.890) and RBF Classifier - Simulated Annealing (Az = 0.924). Set 3, Paraconsistent Classifier (Az = 0.875), MLP classifier (Az = 0.970), ART2LDA Classifier (Az = 0.850) and RBF Classifier - Simulated Annealing (Az = 0.996). Set 4, Paraconsistent Classifier (Az = 0.500), MLP classifier (Az = 0.887), ART2LDA Classifier (Az = 0.767) and RBF Classifier - Simulated Annealing (Az = 0.907). Set 5, Paraconsistent Classifier (Az = 0.929), MLP classifier (Az = 0.987), ART2LDA Classifier (Az = 0.884) and RBF Classifier - Simulated Annealing (Az = 0.998). Set 6, Paraconsistent Classifier (Az = 0.939), MLP classifier (Az = 0.982), ART2LDA Classifier (Az = 0.885) and RBF Classifier - Simulated Annealing (Az = 0.999). In the case of the Paraconsistent Classifier, the eighth experiment was composed of the total image attributes relating to mammographic masses and tumors (Az = 0.939). In the particular case of the RBF-Sort Simulated Annealing experiment with all the image attributes, it was proved its unfeasibility, due to the complexity of their algorithm, where the processing time tends to infinity for a larger number of elements. For the experiments, the Paraconsistent classifier used 20 % of the samples for the neural network training, against the total number of samples available minus one for the other classifiers.
Anzeige
Bitte loggen Sie sich ein, um Zugang zu Ihrer Lizenz zu erhalten.