كاربرد شبكه عصبي مصنوعي در تمايز الگوهاي خوش خيم و بدخيم ضايعات پستاني براساس پارمترهاي فراصوتي

Applying the Artificial Neural Network in Making Discrimination of Bengin and Malignant Patterns of Breast Lesions Using Ultrasonic Parameters

Tissue discrimination using computer aided in diagnosis (CAD) systems through measurement set of physical parameters by ultrasound is an ideal goal. In this study, a neural network based approach was applied to a database consisted of 38 histological tumor samples. This database was consisted of 18 malignant cases (mostly ductal carcinoma) and 18 benign cases (mostly fibro adenoma), which had either biopsy or mastectomy proof. A few comprehensive quantitative features were chosen to discriminate these two classes. These features were attenuation coefficient in 10 MHz in breast benign and malignant lesions at 20, 25, 30 and 35 QC, density and the velocity of the ultrasound in lesions at 30 QC. These findings were obtained from the processing of the ultrasound images of the implemented samples inside a tissue phantom. This database was then normalized between 0 and 1 according to the maximum value of each feature in the data set. The normalized data was then feed into a three layer feed forward neural network with back propagation training algorithm. The jackknife technique used to evaluate the performance of the Artificial Neural Network during the training and testing procedure. The output of the Artificial Neural Network showed a reasonable diagnostic accuracy (75%), a moderate diagnostic specificity (72%) and a high diagnostic sensitivity (77%). These results can be further improved by using a more comprehensive database.