Knowledge Distillation of Teacher-Student Attention Alzheimer's Disease
-
摘要: 通过核磁影像（MRI）分析实现阿尔兹海默症（AD）的早期诊断具有重要意义，但基于深度学习的MRI影像分析通常面对训练样本量小及3D数据计算成本高的挑战，本文提出了一种CNN（Convolutional Neural Networks）和LSTM（Long Short Term Memory）级联的AD诊断模型，先用CNN提取图像的深层特征，再用LSTM对序列特征进行分类，这样将3D的MRI图像视为2D图像的序列，考虑了切片序列间的关联信息。另外，为了提高模型在小样本数据上的性能表现，应用知识蒸馏算法以得到压缩的轻量级模型，并引入师生间的注意力机制提高模型分类的准确率。实验表明，该诊断模型在ADNI数据集上取得了良好的性能。Abstract: In recent years, the detection of Alzheimer's Disease (AD) from neuroimaging data such as MRI by deep learning has become an atractive research method for many scholars. In order to get good model performance on small samples of MRI, this paper proposes a cascade structure consisting of CNN and LSTM for AD diagnosis. Deep features in the image are extracted by the CNN, and then inputted to the LSTM for the classification task. Knowledge distillation is used to obtain compressed lightweight model, and inter-teacher-student attention mechanism is employed to improve the accuracy of model classification. Experiments show that the diagnostic model can achieve good performance on the ADNI dataset.
-
Key words:
- Alzheimer /
- small sample /
- knowledge distillation /
- attention mechanism
-
图 4 AT-KD-den169-den121模型的混淆矩阵
Figure 4. Confusion matrix of the AT-KD-den169-den121 model
表 1 AD的三视图的准确率
Table 1. Accuracy of the three views of AD
View Accuracy/% Sagittal view 82.12 Coronal view 83.83 Axial view 81.70 
下载: 导出CSV
表 2 各模型性能比较
Table 2. Comparison of performance of each model
Model Accuracy/% Precision/% Recall/% F1-Scores/% Teacher-DenseNet169 82.120 — — — Student-Den121 82.128 82.363 82.066 82.035 KD-den169-den121 85.957 85.968 86.042 85.773 AT-KD-den169-den121 86.382 85.710 86.403 85.800
下载: 导出CSV
-
[1] PATTERSON C. The state of the art of dementia research: New frontiers [R]. London: World Alzheimer Report, 2018. [2] FAROOQ A, ANWAR S, AWAIS M, et al. A deep CNN based multi-class classification of Alzheimer's disease using MRI[C]// 2017 IEEE International Conference on Imaging systems and techniques (IST). Beijing: IEEE, 2017: 1-6. [3] 于鲁, 刘卫芳. 海马与胼胝体3D纹理分析在阿尔兹海默症诊断中的比较[J]. 中国医疗设备, 2016, 31(10): 29-32. doi: 10.3969/j.issn.1674-1633.2016.10.009 [4] ZHU Q, ZHANG D, LI H, et al. Incomplete multi-modal brain image fusion for epilepsy classification[J]. Information Sciences, 2022, 582: 316-333. doi: 10.1016/j.ins.2021.09.035 [5] XU M, ZHANG T, ZHANG D. Medrdf: A robust and retrain-less diagnostic framework for medical pretrained models against adversarial attack[J]. IEEE Transactions on Medical Imaging, 2022: 1-1. doi: 10.1109/TMI.2022.3156268 [6] SUBRAMONIAM M. Deep learning based prediction of Alzheimer's disease from magnetic resonance images [EB/OL]. (2021-01-13)[2022-05-21].https://arxiv.org/abs/2101.04961. [7] NAWAZ A, ANWAR S M, LIAQAT R, et al. Deep convolutional neural network based classification of Alzheimer's disease using MRI data[C]// 2020 IEEE 23rd International Multitopic Conference (INMIC). Bahawalpur: IEEE, 2020: 1-6. [8] XING X, LIANG G, BLANTON H, et al. Dynamic image for 3D MRI image Alzheimer’s disease classification[C]// European Conference on Computer Vision. Cham: Springer, 2020: 355-364. [9] KOROLEV S, SAFIULLIN A, BELYAEV M, et al. Residual and plain convolutional neural networks for 3D brain MRI classification[C]// 2017 IEEE 14th international symposium on biomedical imaging (ISBI 2017). Melbourne: IEEE, 2017: 835-838. [10] HINTON G, VINYALS O, DEAN J. Distilling the knowledge in a neural network[J]. Computer Science, 2015, 14(7): 38-39. [11] YANG Y, XUTAO G, YE C, et al. Regularizing brain age prediction via gated knowledge distillation[C]// Proceedings of Medical Imaging with Deep Learning (MIDL), [s. l.]: PMLR 2022: 1430-1443. [12] GUAN H, WANG C, TAO D. MRI-based Alzheimer's disease prediction via distilling the knowledge in multi-modal data[J]. NeuroImage, 2021, 244: 118586. doi: 10.1016/j.neuroimage.2021.118586 [13] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 42(8): 2011-2023. [14] WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C]// Proceedings of the European conference on computer vision (ECCV). [s.l.]: LNIP, 2018: 3-19. [15] PASSBAN P, WU Y, REZAGHOLIZADEH M, et al. ALP-KD: Attention-based layer projection for knowledge distillation[C]// Proceedings of the AAAI Conference on Artificial Intelligence. [s.l.]: AAAI, 2021: 13657-13665. [16] JI M, HEO B, PARK S. Show, attend and distill: Knowledge distillation via attention-based feature matching[C]// Proceedings of the AAAI Conference on Artificial Intelligence. [s.l.]: AAAI, 2021: 7945-7952. [17] YU Y, SI X, HU C, et al. A review of recurrent neural networks: LSTM cells and network architectures[J]. Neural computation, 2019, 31(7): 1235-1270. doi: 10.1162/neco_a_01199 [18] 肖顺, 储呈晨, 王源冰, 等. 压缩感知技术在磁共振成像技术中的应用进展分析[J]. 中国医疗设备, 2021, 36(11): 4-9. [19] HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [s.l.]: IEEE, 2017: 4700-4708. [20] YOSINSKI J, CLUNE J, BENGIO Y, et al. How transferable are features in deep neural networks?[C]// Neural Information Processing Systems. [s.l.]: NIPS, 2014: 1-9. [21] GUO H, ZHENG K, FAN X, et al. Visual attention consistency under image transforms for multi-label image classification[C]// Conference on Computer Vision and Pattern Recognition. Long Beach, CA, USA: IEEE, 2019: 729-739. [22] BELLO I, ZOPH B, VASWANI A, et al. Attention augmented convolutional networks[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 3286-3295. [23] SHENG J, XIN Y, ZHANG Q, et al. Predictive classification of Alzheimer’s disease using brain imaging and genetic data[J]. Scientific Reports, 2022, 12(1): 1-9. doi: 10.1038/s41598-021-99269-x -
点击查看大图
图(4) / 表(2)
计量
- 文章访问数: 205
- HTML全文浏览量: 111
- PDF下载量: 43
- 被引次数: 0
