高级检索

基于迁移学习的特征融合阿尔茨海默病分类算法

Feature Fusion Alzheimer’s Disease Classification Algorithm Based on Transfer Learning

  • 摘要: 为解决阿尔茨海默病(Alzheimer’s disease, AD) MRI图像与中病理特征与正常组织形态差异较小且容易受到无关区域影响分类诊断这一问题, 通过分析MRI图像实现对AD分类的判断及疾病预测, 提出一种基于迁移学习特征融合的AD分类算法. 首先分别从空间域和频域中提取MRI图像的信息特征; 然后利用ADNI数据集解剖视图中的冠状方向的切片, 提高AD二分类任务中的准确率; 再结合改进的融合Vision Transformer和GFNet结构以及迁移学习技术, 提高网络模型的鲁棒性; 最后采用单个患者在冠状方向切片影像数据的多数投票机制, 降低分类过程的偶然性与不确定性, 将AD二分类任务中的准确率提高至0.939. 在公开可用的ADNI数据集与Kaggle数据集上进行对比实验和消融实验, 证明了所提算法在早期诊断AD二分类任务中的潜力与应用价值.

     

    Abstract: In order to solve the problem that the pathological features of Alzheimer’s disease (AD) MRI images are slightly different from normal tissue morphology and are easily affected by irrelevant areas in classifica-tion diagnosis, an AD classification algorithm based on transfer learning feature fusion is proposed by an-alyzing MRI images to realize AD classification judgment and disease prediction. First, the information features of MRI images are extracted from the spatial domain and frequency domain respectively; then, the coronal slices in the anatomical view of the ADNI dataset are used to improve the accuracy of the AD bi-nary classification task; then, the improved fusion Vision Transformer and GFNet structure and transfer learning technology are combined to improve the robustness of the network model; finally, the majority voting mechanism of the coronal slice image data of a single patient is used to reduce the contingency and uncertainty of the classification process, improve the accuracy of AD binary classification task to 0.939. Comparative experiments and ablation experiments are carried out on the publicly available ADNI dataset and Kaggle dataset, which proves the potential and application value of the proposed algorithm in the early diagnosis of AD binary classification tasks.

     

/

返回文章
返回