Brain age – an image derived measure from structural brain images on T1 weighted scans may reveal information on Alzheimer's risk. We have previously shown that increased abdominal adipose tissue relates to brain atrophy. We evaluated the links between abdominal adipose tissue and brain age.
A total of 1,164 healthy participants from four sites (mean chronological age 55.17 ± 12.37 years, 52% women; 48% men; 39% non-white) were scanned on 1.5T MR machines with a whole-body protocol. Whole body sequences utilized in the quantitative analyses of abdominal fat were coronal T1 were used to segment visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) segmentation. In this process, a nnU-Net model was used for fully supervised segmentation and ITK-SNAP was used for manual annotation. Brain age was computed using a regression-based 3D Simple Fully Convolutional Network. The model was trained on in-house T1-weighted MRI scans collected from 5,500 healthy individuals, aged 18 to 89 years. Brain age gap (BAG) was computed by subtracting chronological age from brain age. Bivariate correlations between VAT and SAT to chronological and brain age were done with partial correlations adjusted for sex with brain age. VAT and SAT were normalized to total abdominal body fat volume. Chronological age was not adjusted for in brain age models to avoid collinearity.
Mean brain age exceeded chronological age (mean brain age = 56.04 ± 12.65, mean BAG = 0.69) and were highly correlated (r=0.94, p <.001). VAT and SAT were positively related to increased chronological age (VAT: r=0.2780, p = 5.477e-20; r=0.0924, p = 0.002817) and increased brain age (VAT: r=0.2806, p = 2.42e-20; SAT: r=0.0947, p = 0.002189) with VAT being more closely linked to age than SAT. This did not change when adjusting for sex in separate partial correlations between VAT and SAT for brain age (VAT: rp = r=0.2948, p = 2.247e-22; SAT: r=0.1070, p = 0.0005353). No statistically significant link was noted with VAT, SAT, and BAG.
Both VAT and SAT are linked to chronological and brain age with VAT being more strongly linked. VAT may be a key target for modifying brain age and Alzheimer's risk.
Dementia is a broad category of cognitive decline that affects a person's ability to perform everyday tasks. Unfortunately, there is no cure for dementia, and diagnosis usually happens at later stages when symptoms have significantly progressed. Early detection and accurate staging of dementia could slow down symptom progression and improve the quality of life for affected individuals. In recent years, deep Learning algorithms have emerged as a powerful tool in the early dementia diagnosis. By utilizing MRI images, deep learning techniques can effectively classify dementia at different stages, enabling quicker and more targeted interventions.
We used an MRI dataset from Kaggle, categorized into four classes: non-dementia, very mild dementia, mild dementia, and moderate dementia. The dataset was highly imbalanced, with only 1% of images representing moderate dementia. To address this, we trained a 6-layered Convolutional Neural Network (CNN) using two approaches: (1) a standard 6-layer CNN model and (2) a CNN with SMOTE (Synthetic Minority Over-sampling Technique) to mitigate class imbalance. Model performance was evaluated using accuracy, precision, recall, F1 score, confusion matrix, as well as accuracy and loss curves.
Our SMOTE-CNN model, incorporated the SMOTE technique, showed significant improvements in classifying the four categories of dementia compared to the standard CNN model. The SMOTE-CNN model achieved an overall accuracy of 99%, while the standard CNN model had an accuracy of 71%. Precision, recall, and F1-scores for the SMOTE-CNN model were also notably higher, with perfect scores (1.00) for the 'Mild-Demented' and 'Moderate-Demented' classes, and near-perfect scores (0.99) for the 'Non-Demented' and 'Very Mild-Demented' classes. These results highlight the effectiveness of SMOTE in addressing class imbalance and improving model performance for dementia stage classification.
The results highlight deep learning's potential in enhancing dementia prediction and early detection, leading to better disease staging and management. By addressing class imbalance, we significantly boosted model performance. Future work should focus on validating with independent datasets and incorporating more balanced data to improve generalizability. Additionally, applying the CNN model to real-world data could further enhance dementia prediction and public health outcomes.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：