Feature selection and classification of breast cancer on dynamic magnetic resonance imaging using genetic algorithm and artificial neural networks

الملخص EN

Artificial neural networks (ANN) are frequently used in the development of computer-aide diagnosis systems for breast cancer detection.

Although the ANN may work as an excellent predictor of malignancy, it may not be able to explain which findings are more relevant I reaching the diagnosis.

A hybrid genetic-neural (GA-ANN) model was designed to differentiate malignant from benign in a group of patients with histopathologically proved breast lesions on th base of BI-RADS descriptors and data derived from time-intensity curve.

We used a database with 117 patients' records each of which consisted of 27 quantitative parameters mostly derive from time-intensity curve, 4 BI-RADS qualitative data which determined by expert radiology's and patient age.

These findings were encoded as features for a genetic algorithm (GA) as preprocessor for feature selection and classified with a three-layered neural network to predict th outcome of biopsy.

The network was trained and tested using the jackknife method and it performance was then compared to that of the experienced radiologist in terms of sensitivity specificity, accuracy and receiver operating characteristic curve (ROC) analysis.

The network was able to classify correctly 107 of 117 original cases and yielded a good diagnostic accuracy (91%) sensitivity (95%) and specificity (78%) compared to that of the radiologist (92%), (96% and 78%).

In this paper, GA and ANN techniques were combined for a particular classification problem : The Automatic Prediction of Mammary Biopsy Results from Dynamic MR Imaging.