|
|
1.INTRODUCTIONAlzheimer’s disease is a common senile central neurodegenerative disease and one of the most usual types of dementia. Clinical features include memory deficits, environmental cognitive loss, and motor dysfunction, along with symptoms of psychosis1. Mild cognitive impairment (MCI) is a transitional stage between natural aging and Alzheimer’s disease, and research on anomalous changes in the brain’s white tracts has important implications for the diagnosis and trerapy of cognitive diseases. Due to the rapid progress of noninvasive imaging techniques in recent years, it has become possible to visualize the connection of fiber tracts in the brain. Diffusion tensor imaging (DTI) technology is one of them, which can reconstruct the white matter fiber tracts in the brain2. It also shows how these fiber tracts are connected in the brain. There are two methods commonly used to quantitatively study the features of fibers in the brain: voxel based analysis (VBA) and region based spatial statistics (TBSS)3-5. Since mild cognitive impairment may only cause localized changes in fiber tracts, but not all, the above two methods cannot precisely locate lesions with local abnormalities in fiber tracts. Therefore, this paper takes a deeper look at the measurement of fiber tract indicators within the brain. The automatic fiber bundle quantification algorithm (AFQ) is an open source software (see https://github.com/jyeatman/AFQ), and the working platform is Python. It can automatically identify up to 18 key white matter tracts within the brain and quantify indicators of diffusion characteristics at multiple locations in the fiber tracts6. 2.METHODOLOGY2.1AFQ fiber tractographyThe AFQ method processes can be generalized in four steps:
We use spline interpolation to calculate four fiber bundle diffusion features, namely: fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). We then focus on the FA measure between MCI patients and NC, which is widely used in medical outcome analysis. 2.2FA indicator analysisSPSS 21.0 software was used to perform independent sample t test on the FA indicator of the extracted 18 groups of fiber bundles, and the fiber bundle groups with significant differences between the MCI and NC groups were selected. All subjects information as shown in Table 1. The statistical results of fibers are as shown in Table 2. Table 1.Information of MCI patients and NC.
Table 2.Comparison of FA between MCI and NC groups.
3.DATA3.1Image data acquisitionThe parameters of the NMR data we used are as belows: repetition time (TR) = 3400 ms; echo time (TE) = 100 ms; gap 0 mm; flip angle = 135°; field of view (FOV) = 220× 220 mm2; acquisition matrix = 300×190. Imaging sequence parameters for DTI images are as belows: TR = 6600 ms; TE = 90 ms; number of slices 45; slice thickness 3.0 mm; gap 0 mm; b-value = 1000 s/mm2; = 128×128. 3.2SubjectsFrom January 2019 to December 2020, 29 patients with confirmed MCI were recruited by the Brain Hospital Affiliated to Nanjing Medical University, including 22 males and 16 females, aged (71.96±8.23) years. Meanwhile, 38 healthy elderly people matched with the patient group in gender, age and years of education were recruited as healthy control group. Each of the above subjects underwent whole-brain MRI scans, physical examinations, and neuropsychological scales such as Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Clinical Dementia Rating Scale (CDR) scores and so on (Table 1). 4.RESULTS4.1Mean diffusion measures of MCI and NC groupsIn Table 2, we list the number of subjects in each tract (named n1:n2) that were correctly tracked using FDR correction. The AFQ method does not perform well in some fibers, especially the large tract of corpus callosum in this study. Fiber bundles like CGC_R, CST_L and IFOF_R are missing 1-2 FA values, and the tracking results of other fiber bundles (such as ILF_R, SLF_R and UNC_R) are complete. This may be caused by the threshold setted of the fiber tracking, or it may be due to individual differences or atrophy of the associated fiber bundles. The total number of fiber tracts traced to MCI and NC will be added to future studies to better assess white matter damage in the brain. We used a p < 0.05 threshold to compare the FA feature of each white matter tract between the MCI and NC groups in a statistical analysis and identified tracts with significant differences. Compared with the normal controls, four fiber tracts showed a significant difference, which were bolded: CST_R, IFOF_L, UNC_R and FP. The magnitude of the FA indicator value was also significantly different between the MCI and NC groups. Figures 1-4 show the FA indicator of some fiber bundles between the two groups. The horizontal axis is the number of subjects (38 cases in total), and the vertical axis is the FA index value. The blue line segment is the NC group, and the orange line is the MCI group. 4.2SVM based classificationFigure 5 is the classification result, where 1 is a predicted positive sample, 0 is a negative sample. An error value of 0 indicates that the classification is correct, 1 is a correct sample that is judged to be an incorrect sample, -1 is an incorrect sample that is judged to be a correct sample, and the overall value is subtracted 5, which is -5, -4 and -6 to avoid confusion. The true negative (TN), true positive (TP), false positive (FP) and false negative (FN) results are 35, 34, 3 and 4 seperately, so the accuracy, sensitivity and specificity are 90.78% (69/76), 89.47% (34/38) and 92.11% (35/38)8-10. 5.CONCLUSIONBased on the AFQ method, this paper proposes a classification and recognition algorithm for MCI and NC groups, and the final recognition rate is good, which shows that this method is feasible and effective. At the same time, judging from the results of significant differences between MCI and NC groups of fibers, there is a statistically significant difference in the distribution of commissural fibers, commissural fibers and projection fibers in the brain, and further research is needed to explain this in the future. Different from studies using MR image texture features, hippocampal volume features, etc., this paper innovatively uses fiber bundle groups with significant differences to train the SVM model, and obtains high accuracy, sensitivity and specificity, which illustrates the method in this paper. It is effective. Future work will consider including the gender, age, etc. of the subjects into the features, and consider combining multimodal data to construct feature vectors to increase the accuracy of the classification method. REFERENCESJin, L. and Guo, Q.,
“Research progress in neuroimaging of Alzheimer’s disease,”
J. Diagn. Concepts Pract, 6
(1), 66
–70 Google Scholar
Yang, W., Zou, H. and Hu, X.,
“Preliminary study of DTI in the evaluation of changes of white matter microstructure in patients with depression,”
Chinese Journal of CT and MRI, 18
(12), 1
–6
(2020). Google Scholar
Huang, J., Guo, D., Zhang, B., et al.,
“Study on the microstructure of white matter in patients with postpartum depression,”
China J. Nerv. Ment. Dis, 46
(10), 119
–29
(2020). Google Scholar
Yu, T., Yin, Z., Yi, X., et al.,
“White matter integrity in patients with mild cognitive impairment complicated with lacunar infarctions by diffusion tensor imaging,”
J. Cent. South. Univ. (Med. Sci.), 44
(7), 805
–12
(2019). Google Scholar
Li, W., Wang, F., Zhang, X., Li, M., Lu, J. M., Wu, S. C. and Zhang, B.,
“Study on age and white matter neuronal integrity in healthy volunteers based on automating fiber-tract quantification,”
National Medical Journal of China, 97
(13), 976
–81
(2011). Google Scholar
Yeatman, J. D., Dougherty, R. F., Myall, N. J., Wandell, B. A. and Feldman, H. M.,
“Tract profiles of white matter properties: Automating fiber-tract quantification,”
PLoS ONE, 7
(11), e49790
(2012). https://doi.org/10.1371/journal.pone.0049790 Google Scholar
Zhang, X., Sun, Y., Li, W., et al.,
“Characterization of white matter changes along fibers by automated fiber quantification in the early stages of Alzheimer’s disease,”
NeuroImage: Clinical, 22
(2019). Google Scholar
Hui, L., Jie, X., Jia, Q., et al.,
“Classification of mild cognitive impairment and Alzheimer disease based on Adabooast algorithm,”
Chin J. Med. Imaging. Technol, 32
(4), 623
–627
(2016). Google Scholar
Xing, Z. and Wei, F.,
“Classification studies in patients with mild cognitive impairment and normal control,”
Medical Engineering Technology, 32
(10), 76
–79
(2017). Google Scholar
Zhang, D., Wang, Y. and Zhou, L.,
“Multimodal classification of Alzheimer’s disease and mild cognitive impairment,”
NeuroImage, 55
(3), 856
–857
(2011). https://doi.org/10.1016/j.neuroimage.2011.01.008 Google Scholar
|
