Brainstem texture analysis can provide valuable information in the diagnosis of Early Mild Cognitive Impairment (EMCI) condition. In this work, 3D brainstem structure is segmented and analysed for texture alterations using multifractal features to differentiate EMCI from other Alzheimer’s disease stages. The images obtained from public domain database are preprocessed for spatial registration, skull stripping and contrast enhancement. White matter volume is segmented from the preprocessed images using fuzzy ‘C’ means clustering algorithm. Midsagittal white matter tissue is used as the initial seed to segment the brainstem volume using sparse field level set method. Multifractal detrended moving average algorithm is used to compute the fluctuation function, generalized Hurst exponent and mass exponent to study the multifractal characteristics of brainstem structure. Features extracted from the multifractal spectrum are analysed to differentiate the images pertaining to EMCI subject group. Results indicate that the proposed technique is able to segment the brainstem structure from all the considered images. The fluctuation function is observed to have linear relationship with scale. The generalized Hurst exponent decreases with order and mass exponent follows a non-linear trend demonstrating the multifractal nature of brainstem. Singularity spectral features namely strength of multifractality, Holder exponent at f(2.8), tangent slope and maximum Holder exponent are found to be most significant in differentiating EMCI from subject groups. As this complex EMCI distinction is clinically important, the proposed approach is useful for early diagnosis of Alzheimer’s condition.
Read full abstract