Learning the similarity in breast cancer tissue images
Xin Wang · International Journal of Healthcare Technology and Management · 2006
This paper is to establish a system that is able to understand and recognise similarity patterns in medical image collection in terms of local low-level primitives and global semantics. The experimental data used by the system are histological tissue images on breast cancer. The image colour features are defined in spatial domain at pixel level and converted into frequency domain by Fourier transform. The transformed data, therefore, can be compressed at different degrees effectively. Once the image features are compressed and divided into small blocks, the neural network of the system will be trained that each block is regarded as a pattern in the patterns database, which can be compared and matched by a new query image. The prototype has proved interesting for comparison of medical images, and can be used for training new doctors and providing reference for diagnosis. The method is of generic nature and can be applied to other domains.