Discriminant analysis of background noise in extremity magnetic resonance images
Resumo
Since the creation of the first magnetic resonance imaging (MRI) equipment in 1974, experts have been studying the continuous improvement of image quality. This work aims to study the types of background noise in images from extremity MRI system of high-field, mainly caused by Faraday Cage problems. Phantom images of 1T equipment were investigated for this study. For the acquisition of these images, a protocol called DQA (Daily Quality Assurance) was used. For this work, 45 MRI images were acquired, which were pre-classified by an expert, and analyzed by SNR, an index that quantifies the ratio between signal and image noise, and by the multivariate statistical methods PCA + MLDA. PCA served as a statistical filter, which considerably decreased the amount of input information for MLDA. When all main components were used, MLDA showed an accuracy of 93.33% and results that allowed to discriminate background noise from these images in complementarity with SNR.
Referências
Fisher, R. A. (1936). The use of multiple measurements in taxonimic problems. Annals of Eugenics, 7(2):179-188.
Fukunaga, K. (1990). Introduction to Statistical Pattern Recognition. Cambridge University Press.
Giraldi, G., Thomaz, C., and Rodrigues, P. (2008). Dimensionality reduction, classification and recosnstruction problems in statistical learning approaches. Revista de Informatica Teórica e Aplicada.
Goerner, F. L. and Clarke, G. D. (2011). Measuring signal to noise ratio in partially paralell imaging mri. Medical Physics, 38:5049-5057.
Guerin, B., Villena, J. F., Polimeridis, A. G., Adalsteinsson, E., Daniel, L., White, J. K., and Wald, L. L. (2017). The ultimate signal-to-noise ratio in realistic body models. Magnetic Ressonance in Medicine, 78:1969-1980.
Hastie, T., Tibshirani, R., and Friedman, J. (2009). The Elements of Statistical Learning:Data Mining, Inference and Prediction. Springer.
Janousova, E., Schwarz, D., and Kasparek, T. (2015). Combining various types of classifiers and features extracted from magnetic resonance imaging data in schizophrenia recognition. Psychiatry Research: Neuroimaging, 232:237-249.
Jin, N. r., Saybasili, H., and Bi, X. (2015). Sparse coding for improved signal-to-noise ratio in mri. Basic Principles of Cardiovascular MRI, pages 41-62.
Koriwakova, E., Schwarz, D., and Kaparek, T. (2010). Classification of 3-d mri brain data using modified maximum uncertainty linear discriminant analysis.
Kuperman, V. (2000). Magnetic Ressonance Imaging. Academic Press.
Marcovski, A. (1996). Noise in mri. Magnetic Ressonance in Medicine, 36:494-497.
McRobbie, D., Moore, E., Graves, M., and Prince, M. (2006). Mri from picture to proton. Academic Press.
Oni (2003). Optimizing image quality.
Pang, X. and Mak, M.-W. (2015). Noise robust speaker verification via the fusion of snr-independent and snr-dependent plda. International Journal of Speech Technology, 18:633-648.
Sato, J. R., Fujita, A., Thomaz, C. E.and Martin, M. G. M., Mourão, M. J., Brammer, M. J., and Junior, E. A. (2009). Evaluating svm and mlda in the extraction of discriminant regions for mental state prediction. Neuroimage, 46(1):105-114.
Sharma, N. (2017). Single-trial p300 classification using pca with lda, qda and neural networks. ArXiv, abs/1712.01977.
Thomaz, C.E.and Kitani, C. and Gillies, D. (2006). A maximum uncertainty lda-based approach for limited sample size problems - with application to face recognition. Journal of the Brazilian Computer Society (JBCS), 12(2):7-18.
Thomaz, C.E.and Kitani, E., P.S., R., Girald, G., Rueckert, D., and Gillies, D. (2006). Discriminant principal components. Journal of the Brazilian Computer Society (JBCS), 1(2):1-12.
