Biomedical Scan Image Retrieval Using Higher Order Neural Networks

The automatic medical diagnosis that uses computer tomography or magnetic resonance images or any other medical images requires retrieving most similar images from an image repository of previous patients when an image of a particular patient is given as a query. In the existing methods, the researchers have used local bit plane decoded pattern (LBDP), local diagonal extrema pattern (LDEP), local binary pattern (LBP), and local wavelet pattern (LWP) for indexing and retrieval of biomedical images. In this proposed method, the pulse-coupled neural networks (PCNN) and unit-linking PCNN are used to extract the features. The PCNN is a single-layer two-dimensional neural network of pulse-coupled neurons. Each pixel in the image is feeding input to the corresponding neuron in the 2D network. Thus, the feature extraction is done using basic and UL-PCNN. Then to retrieve similar images, the classification is done using the neural networks. The proposed algorithm is tested for image retrieval using the mammographic image analysis society (MIAS) database and the existing approaches.