Prediction of 7‐year's conversion from subjective cognitive decline to mild cognitive impairment

Ling Yue,Dan Hu,Han Zhang,Junhao Wen,Ye Wu,Wei Li,Lin Sun,Xia Li,Jinghua Wang,Guanjun Li,Tao Wang,Dinggang Shen,Shifu Xiao
DOI: https://doi.org/10.1002/hbm.25216
IF: 4.8
2020-10-08
Human Brain Mapping
Abstract:<p>Subjective cognitive decline (SCD) is a high‐risk yet less understood status before developing Alzheimer's disease (AD). This work included 76 SCD individuals with two (baseline and 7 years later) neuropsychological evaluations and a baseline T1‐weighted structural MRI. A machine learning‐based model was trained based on 198 baseline neuroimaging (morphometric) features and a battery of 25 clinical measurements to discriminate 24 progressive SCDs who converted to mild cognitive impairment (MCI) at follow‐up from 52 stable SCDs. The SCD progression was satisfactorily predicted with the combined features. A history of stroke, a low education level, a low baseline MoCA score, a shrunk left amygdala, and enlarged white matter at the banks of the right superior temporal sulcus were found to favor the progression. This is to date the largest retrospective study of SCD‐to‐MCI conversion with the longest follow‐up, suggesting predictable far‐future cognitive decline for the risky populations with baseline measures only. These findings provide valuable knowledge to the future neuropathological studies of AD in its prodromal phase.</p>
radiology, nuclear medicine & medical imaging,neurosciences,neuroimaging
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to predict whether subjective cognitive decline (SCD) will progress to mild cognitive impairment (MCI) within 7 years. Specifically, researchers hope to use machine - learning models to distinguish individuals with SCD who will eventually develop into MCI (i.e., progressive SCD, abbreviated as pSCD) from those with SCD whose cognitive status remains stable (i.e., stable SCD, abbreviated as sSCD) by combining neuroimaging features and clinical measurement data at baseline. The research on this problem has important clinical significance because it can help in the early identification of high - risk populations who may develop Alzheimer's disease (AD), thus providing a basis for early intervention. ### Research Background 1. **Importance of SCD**: SCD refers to the subjective perception of cognitive decline by an individual, but no obvious deficits have been shown in objective neuropsychological tests. Research indicates that SCD may be an important sign in the early stage of AD, especially the association between memory function problems and AD is the most significant. 2. **Limitations of Existing Research**: - Most existing longitudinal studies have a relatively short follow - up period (2 - 5 years), which is not sufficient to accurately predict long - term cognitive decline. - Many previous studies mainly relied on cross - sectional designs and it was difficult to obtain conclusive results. - Invasive detection methods such as PET or CSF can detect Aβ or tau - related biomarkers, but they are costly and not suitable for large - scale screening. ### Research Objectives This study aims to construct a model that can predict whether SCD will progress to MCI within 7 years by using machine - learning methods and non - invasive structural magnetic resonance imaging (sMRI) and clinical/demographic features. Specific objectives include: 1. **Identifying Predictors**: Determine which baseline features (such as neuroimaging indicators, education level, past medical history, etc.) can effectively predict the transformation of SCD to MCI. 2. **Validating the Effectiveness of the Model**: Validate the effectiveness of the identified predictors and their combinations through an independent data set. 3. **Providing Clinical Application Value**: Develop a simple, automated prediction model to help doctors conduct early risk assessment in clinical practice. ### Research Methods 1. **Research Cohort**: Use data from the China Longitudinal Aging Study of Cognitive Impairment (CLAS), including 76 individuals with SCD, of which 24 progressed to MCI after 7 years and 52 remained stable. 2. **Feature Extraction**: Extract 198 neuroimaging features from T1 - weighted structural MRI and collect 25 clinical measurement indicators (such as years of education, MoCA score, etc.). 3. **Feature Selection and Model Training**: Adopt a three - stage feature selection scheme (Relief algorithm, comprehensive ranking, cost - sensitive support vector machine - based classification performance evaluation), and finally determine five most contributing features. 4. **Model Validation**: Validate the prediction performance of the model through nested leave - one - out cross - validation (LOOCV). ### Main Findings 1. **Key Predictors**: - A history of stroke at baseline - Lower years of education - Lower baseline MoCA score - Reduced volume of the left amygdala - Increased white matter volume in the right superior temporal sulcus 2. **Model Performance**: Based on these five features, the model has good prediction performance, with an accuracy rate of 69.74%, a sensitivity of 62.50%, a specificity of 73.08%, and an F1 score of 0.5660. In particular, through the simplified five - feature model, the area under the ROC curve (AUC) reached 0.7997. 3. **Relationship between MRI Features and Cognitive Scores**: The volume of the left amygdala is negatively correlated with the subsequent MoCA score (r = - 0.636, p = 0.003), indicating that the reduction in the volume of the left amygdala is associated with cognitive function decline.