Wavelet based independent component analysis for multispectral brain tissue classification

Abstract

Multispectral analysis is a promising approach in\ntissue classification and abnormality detection from Magnetic\nResonance (MR) images. But instability in accuracy and\nreproducibility of the classification results from conventional\ntechniques keeps it far from clinical applications. Recent studies\nproposed Independent Component Analysis (ICA) as an effective\nmethod for source signals separation from multispectral MR data.\nHowever, it often fails to extract the local features like small\nabnormalities, especially from dependent real data. A multisignal\nwavelet analysis prior to ICA is proposed in this work to resolve\nthese issues. Best de-correlated detail coefficients are combined\nwith input images to give better classification results.\nPerformance improvement of the proposed method over\nconventional ICA is effectively demonstrated by segmentation\nand classification using k-means clustering. Experimental results\nfrom synthetic and real data strongly confirm the positive effect\nof the new method with an improved Tanimoto index/Sensitivity\nvalues, 0.884/93.605, for reproduced small white matter lesions