Feature fused breast cancer detection
K.P. Adila, K. Sheeba · 2020
Breast cancer is dreadful, fatal and a widespread disease now. Mammogram is a commonly used imaging technique for screening. Deep learning is applied in clinical dataset for medical imaging. Images are affected by quality factors. To diagnose more accurately pre-processing methods are required. Here denoising is done by wiener filter and image enhancement is performed by contrast limited adaptive histogram equalization. As medical image segmentation has an important role in computer aided diagnosis, a hybrid segmentation technique, thresholding and morphological operation is done. Image feature analysis is done by a convolutional neural network and geometric and texture feature analysis is done by multi-layer perceptron. A multi-modal feature fusion technique which merge low-level features with high-level features is applied. Late multimodal fusion technique implemented combines the output from individual classifiers and differentiates benign and malignant growth. In the performed work AlexNet architecture shows an improved accuracy in classifying.