There are many active research projects accessing and applying shared ADNI data. Use the search above to find specific research focuses on the active ADNI investigations. This information is requested annually as a requirement for data access.
| Principal Investigator | |
| Principal Investigator's Name: | zhenyu shu |
| Institution: | Zhejiang Provincial People's Hospital Affiliated to Hangzhou Medical College |
| Department: | Radiology Department |
| Country: | |
| Proposed Analysis: | For the current study, we selected 357 patients with MCI for whom datasets were available, including baseline MRI, genetic and neuropsychological data, who had at least 3 years of clinical follow-up. A total of 154 patients converted to AD during the 3-year follow-up period. In addition, according to the original ADNI participant number, the study cases were divided into a training set (n = 249) and a test set (n=108) according to a 7:3 ratio. The prediction model was established by the training set, and the reliability of the model was verified by the test set. We obtained T1WI MRI data from the ADNI database, which were acquired with a 1.5 T scanner by a standardized MPRAGE protocol. In addition to MRI measures, non-invasive, inexpensive and easy-to-obtain neuropsychological measures were also employed. All the subjects completed the comprehensive neuropsychological scales, including the Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR), reflecting the overall cognitive status, and the Alzheimer's Disease Assessment Scale (ADAS), reflecting the disease condition. The details of the test procedures and scoring criteria are given in the ADNI General Procedures Manual(21). In addition, genetic data were included as a demographic variable for further study. 1.3 Image preprocessing We imported the T1WI images into the SPM12 software (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) in the DICOM format for automatic segmentation of the whole-brain WM, GM and CSF. The WM, GM and CSF volumes were manually modified by two experienced neuroradiologists (radiologist A and radiologist B, with 6 and 10 years of neuroimaging experience, respectively), who were blinded to the clinical data, using ITK-SNAP software (http://www.itksnap.org). This modification was accomplished according to the following steps: (1) the removal of non-brain tissue, brainstem, and cerebellum and (2) the correction of segmentation errors in brain tissue. The brain tissue after manual correction was then imported into QK software (Quantitative Analysis Kit, version 1.2, GE Healthcare) for feature extraction. In addition, image preprocessing was applied before feature extraction. First, T1WIs were resampled at a single-voxel resolution of 1×1×1 mm3 by linear interpolation. The image greyscale intensity level was then discretized and normalized by downsampling each image into 32 bins to reduce image noise. The IPM software package of the QK analysis platform was used to extract the radiomics features, including the histogram, Haralick texture features, grey level co-occurrence matrix (GLCM), run length matrix (RLM) and grey level size zone matrix (GLZSM) features. In addition, we used more robust features after manual segmentation by two radiologists to ensure the reproducibility and repeatability of the radiomics features. The Spearman rank correlation test was used to calculate the correlation coefficient between feature set A (from radiologist A) and feature set B (from radiologist b). According to the rule of thumb, features with a correlation coefficient greater than 0.8 are considered robust features. In addition, to prevent the "curse of dimension" caused by too many features to make the prediction result inaccurate, we reduced the dimension of the extracted features to reduce the number of features. We first used the maximum relevance minimum redundancy (mRMR) algorithm to reduce the dimension of the extracted robust features. The mRMR algorithm aims to select a group of features, which can be divided into two categories (stable and transformed) to the greatest extent. At the same time, the intracorrelation of features is minimized. Finally, we use the traditional least absolute shrinkage and selection operator (LASSO) algorithm to further reduce the dimensions of the selected features to generate a final set of top-level radiation features related to MCI transformation and to participate in the construction of the radiomics signature. |
| Additional Investigators |
