Breast Cancer Detection and Classification using Global Pooling

Breast cancer is one of the leading causes of cancer related death in women. Studies suggest that early detection can bring down the mortality rate to a great extent. Out of different imaging techniques, doctors and radiologists prefer mammography for early detection of breast cancer and so mammography remains the gold standard screening technique. However, due to overwork and lack of expertise,a lot of misclassification is also reported. Computer Aided Diagnosis has helped doctors and radiologists in this regard. With the rapid development of the Convolutional Neural Network image classification has reported major success. Our work uses CNN for classification of ROIs extracted from the mammograms. A novel technique for feature extraction from different convolutional layers of the DCNN using Global Average Pooling and later concatenating all the extracted features before the final classification are proposed. We have experimentally proved that incorporating such techniques can improve classification performance in almost all the pre-trained DCNN structures.