基于序列缺失的MRI多序列特征填补与融合互助模型：鉴别高低级别胶质瘤

doi:10.12122/j.issn.1673-4254.2024.08.15

南方医科大学学报 ›› 2024, Vol. 44 ›› Issue (8): 1561-1570.doi: 10.12122/j.issn.1673-4254.2024.08.15

• • 上一篇

基于序列缺失的MRI多序列特征填补与融合互助模型：鉴别高低级别胶质瘤

吴垂杏¹(), 钟伟雄¹, 谢金城¹, 杨蕊梦²^,³, 吴元魁⁴, 许乙凯⁴, 王琳婧⁵, 甄鑫¹()

^1.南方医科大学生物医学工程学院，广东广州 510515
^2.广州市第一人民医院放射科，广东广州 510180
^3.华南理工大学医学院，广东广州 510006
^4.南方医科大学南方医院医学影像科，广东广州 510515
^5.广州医科大学附属肿瘤医院，广东广州 510095

收稿日期:2024-04-19 出版日期:2024-08-20 发布日期:2024-09-06
通讯作者: 甄鑫 E-mail:1527936022@qq.com;xinzhen@smu.edu.cn
作者简介:吴垂杏，硕士，E-mail: 1527936022@qq.com
基金资助:
国家自然科学基金(82371908);国家自然科学基金青年基金(62106058);广东省自然科学基金(2022A1515011410)

An MRI multi-sequence feature imputation and fusion mutual-aid model based on sequence deletion for differentiation of high-grade from low-grade glioma

Chuixing WU¹(), Weixiong ZHONG¹, Jincheng XIE¹, Ruimeng YANG²^,³, Yuankui WU⁴, Yikai XU⁴, Linjing WANG⁵, Xin ZHEN¹()

^1.School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
^2.Department of Radiology, Guangzhou First People's Hospital, Guangzhou 510180, China
^3.School of Medicine, South China University of Technology, Guangzhou 510006, China
^4.Department of Medical Imaging, Nanfang Hospital, Southern Medical University, Guangzhou 510515, China
^5.Guangzhou Institute of Cancer Research, the Affiliated Cancer Hospital, Guangzhou Medical University, Guangzhou 510095, China

Received:2024-04-19 Online:2024-08-20 Published:2024-09-06
Contact: Xin ZHEN E-mail:1527936022@qq.com;xinzhen@smu.edu.cn
Supported by:
National Natural Science Foundation of China(82371908)

摘要/Abstract

摘要：

目的探讨基于序列缺失的MRI多序列特征填补与融合互助模型应用于高级别胶质瘤（HGG）与低级别胶质瘤（LGG）鉴别的性能表现。方法回顾性收集305例胶质瘤患者（189例HGG，116例LGG）的MRI图像，分别勾画出T1加权成像（T1WI）、T2加权成像（T2WI）、T2液体翻转恢复衰减（T2_FLAIR）和T1WI增强图像（CE_T1WI）的感兴趣区（ROI），提取出4个ROI的影像组学特征。利用本研究提出的基于序列缺失的MRI多序列特征填补与融合互助模型对含有缺失数据的特征矩阵进行填补与融合双向学习得到互助模型。采用五折交叉验证方法和准确率（ACC）、平衡准确率（BAcc）、ROC曲线下的面积（AUC）、特异性和灵敏度评价该模型的鉴别能力。所提模型与其他非完整多模态分类模型在鉴别HGG与LGG上进行定量比较，对本文提出的特征填补与融合方法学习得到的潜在特征进行类可分性实验，观察样本在二维平面的分类效果，采用收敛性实验验证该模型的可行性。结果模型序列缺失率为10%时，其在鉴别HGG与LGG的ACC、BAcc、AUC、特异性、灵敏度分别为：0.777、0.768、0.826、0.754和0.780，融合的潜在特征在类可分性实验中有优秀表现，该算法可迭代至收敛。缺失率为30%、50%时，分类性能也优于其他方法。结论基于序列缺失的MRI多序列特征填补与融合互助模型在HGG和LGG的分类任务中具有优异的性能表现。与其他非完整多模态分类模型相比，该模型在鉴别HGG和LGG的分类性能更优，适用于非完整模态的多模态数据的处理。

关键词: 序列缺失, 特征填补, 表征学习, 高级别胶质瘤, 低级别胶质瘤

Abstract:

Objective To evaluate the performance of magnetic resonance imaging (MRI) multi-sequence feature imputation and fusion mutual model based on sequence deletion in differentiating high-grade glioma (HGG) from low-grade glioma (LGG). Methods We retrospectively collected multi-sequence MR images from 305 glioma patients, including 189 HGG patients and 116 LGG patients. The region of interest (ROI) of T1-weighted images (T1WI), T2-weighted images (T2WI), T2 fluid attenuated inversion recovery (T2_FLAIR) and post-contrast enhancement T1WI (CE_T1WI) were delineated to extract the radiomics features. A mutual-aid model of MRI multi-sequence feature imputation and fusion based on sequence deletion was used for imputation and fusion of the feature matrix with missing data. The discriminative ability of the model was evaluated using 5-fold cross-validation method and by assessing the accuracy, balanced accuracy, area under the ROC curve (AUC), specificity, and sensitivity. The proposed model was quantitatively compared with other non-holonomic multimodal classification models for discriminating HGG and LGG. Class separability experiments were performed on the latent features learned by the proposed feature imputation and fusion methods to observe the classification effect of the samples in two-dimensional plane. Convergence experiments were used to verify the feasibility of the model. Results For differentiation of HGG from LGG with a missing rate of 10%, the proposed model achieved accuracy, balanced accuracy, AUC, specificity, and sensitivity of 0.777, 0.768, 0.826, 0.754 and 0.780, respectively. The fused latent features showed excellent performance in the class separability experiment, and the algorithm could be iterated to convergence with superior classification performance over other methods at the missing rates of 30% and 50%. Conclusion The proposed model has excellent performance in classification task of HGG and LGG and outperforms other non-holonomic multimodal classification models, demonstrating its potential for efficient processing of non-holonomic multimodal data.

Key words: sequence deletion, feature imputation, representation learning, high-grade glioma, low-grade glioma

吴垂杏, 钟伟雄, 谢金城, 杨蕊梦, 吴元魁, 许乙凯, 王琳婧, 甄鑫. 基于序列缺失的MRI多序列特征填补与融合互助模型：鉴别高低级别胶质瘤[J]. 南方医科大学学报, 2024, 44(8): 1561-1570.

Chuixing WU, Weixiong ZHONG, Jincheng XIE, Ruimeng YANG, Yuankui WU, Yikai XU, Linjing WANG, Xin ZHEN. An MRI multi-sequence feature imputation and fusion mutual-aid model based on sequence deletion for differentiation of high-grade from low-grade glioma[J]. Journal of Southern Medical University, 2024, 44(8): 1561-1570.

图/表 10

表1 HGG和LGG患者队列的人口统计学特征

Tab.1 Demographic characteristics of the patients with HGG and LGG ［n (%)］

Characteristics	HGG(n=189)	LGG(n=116)	$χ 2$	P
Age (year)
≤50	108 (57.1%)	107(92.2%)	-7.021*	2.460
>50	81 (42.9%)	9(7.8%)	-7.021*	2.460
Gender
Female	73 (38.6%)	48(41.4%)	1.652^#	0.199
Male	116 (61.4%)	68(58.6%)	1.652^#	0.199

表1 HGG和LGG患者队列的人口统计学特征

Tab.1 Demographic characteristics of the patients with HGG and LGG ［n (%)］

Characteristics	HGG(n=189)	LGG(n=116)	$χ 2$	P
Age (year)
≤50	108 (57.1%)	107(92.2%)	-7.021*	2.460
>50	81 (42.9%)	9(7.8%)	-7.021*	2.460
Gender
Female	73 (38.6%)	48(41.4%)	1.652^#	0.199
Male	116 (61.4%)	68(58.6%)	1.652^#	0.199

图1 基于序列缺失的MRI多序列特征填补与融合互助模型流程图

Fig.1 Framework of MRI multi-sequence feature imputation and fusion mutual aid model based on sequence deletion.

表2 所提取的MRI影像组学特征

Tab.2 Extracted MRI radiomics features

Categories of radiomics features	Radiomics features
First Order Features (m=19)	Mean, Median, Maximum, Minimum, Robust Mean Absolute Deviation, Root Mean Squared, Mean Absolute Deviation, 10th percentile, 90th percentile, Uniformity, Standard Deviation, Skewness, Kurtosis, Variance, Energy, Total Energy, Entropy Minimum, Range, Interquartile Range
Shape Features (m=15)	Elongation, Flatness, Major Axis Length, Least Axis Length, Maximum 2D diameter (Column), Maximum 2Ddiameter (Row), Maximum 2D diameter (Slice), Maximum 3D diameter, Minor Axis Length, Sphericity, Surface Area, Mesh Volume, Surface Volume Ratio, Spherical Disproportion, Voxel Volume
Texture Features (m=75)	¹NGTDM (m=5), ²GLDM (m=14), ³GLSZM (m=16), ⁴GLRLM (m =16), ⁵GLCM (m =24)

表3 基于序列缺失的MRI多序列特征填补与融合互助算法伪代码

Tab.3 Pseudocode of MRI multi-sequence feature imputation and fusion mutual aid model based on sequence deletion

Algorithm 1 Pseudocode of the Proposed Method

Training stage

Input: multimodal feature matrix $X = X 1; …; X i; …; X D, i ∈ 1, D$ ,feature dimension $m i$ and hyperparameters $k, λ i$ and $γ i$ .

Output: $V$ , $P$

Begin

Initialize $Q, V, P, H$ as random matrices;

For $t = 1$ to $T$ do

Update $Q$ by using $Q - α 1 ∇ Φ Q$

Update $V$ by using $V - α 2 ∇ Φ V$

Update $P$ column by column by using $p j = Z f j T f j f j T + γ 3 / p j 2 - 1$

Update $H$ by using $H - α 3 ∇ Φ H$

End

Testing stage

Input: new unseen multimodal feature $X = X 1; …; X i; …; X D, i ∈ 1, D$

Output: fused representation $V n e w = P X ∘ U$ , where $U ∈ R k × n$ , $u i j = 1 / D j$ , $i ∈ 1, k, j ∈ 1, n,$ $D j$ is the number of complete modalities in the $j t h$ sample.

表3 基于序列缺失的MRI多序列特征填补与融合互助算法伪代码

Tab.3 Pseudocode of MRI multi-sequence feature imputation and fusion mutual aid model based on sequence deletion

Algorithm 1 Pseudocode of the Proposed Method

Training stage

Input: multimodal feature matrix $X = X 1; …; X i; …; X D, i ∈ 1, D$ ,feature dimension $m i$ and hyperparameters $k, λ i$ and $γ i$ .

Output: $V$ , $P$

Begin

Initialize $Q, V, P, H$ as random matrices;

For $t = 1$ to $T$ do

Update $Q$ by using $Q - α 1 ∇ Φ Q$

Update $V$ by using $V - α 2 ∇ Φ V$

Update $P$ column by column by using $p j = Z f j T f j f j T + γ 3 / p j 2 - 1$

Update $H$ by using $H - α 3 ∇ Φ H$

End

Testing stage

Input: new unseen multimodal feature $X = X 1; …; X i; …; X D, i ∈ 1, D$

Output: fused representation $V n e w = P X ∘ U$ , where $U ∈ R k × n$ , $u i j = 1 / D j$ , $i ∈ 1, k, j ∈ 1, n,$ $D j$ is the number of complete modalities in the $j t h$ sample.

表4 混淆矩阵

Tab.4 Confusion Matrix

Actual\\Predicted	Predicted Negative (PN)	Predicted Positive (PP)
Actual Negative (AN)	True Negative (TN)	False Positive (FP)
Actual Positive (AP)	False Negative (FN)	True Positive (TP)

表5 将融合多模态数据输入不同分类器的性能比较

Tab.5 Performance comparison after feeding the fusion multimodal data into various classifiers

Classifier	ACC/BAcc	AUC	SPE	SEN
SVM	0.770/0.758	0.824	0.724	0.792
LDA	0.761/0.743	0.797	0.684	0.802
KNN	0.767/0.751	0.761	0.709	0.792
Bagging	0.744/0.731	0.790	0.689	0.772
XGBoost	0.764/0.743	0.794	0.668	0.819
AdaBoost	0.731/0.711	0.711	0.651	0.771
Gaussian NB	0.767/0.756	0.790	0.737	0.774
Logistic regression	0.777/0.768	0.826	0.754	0.780

表6 本模型与其他模型在不同缺失率下对HGG与LGG的分类性能比较

Tab.6 Classification performance of this model and other models for discriminating HGG from LGG at different missing rates

Model	10%			30%			50%
Model	ACC/BAcc	SPE	SEN	ACC/BAcc	SPE	SEN	ACC/BAcc	SPE	SEN
BSV	0.639/0.611	0.491	0.730	0.662/0.613	0.405	0.820	0.646/0.569	0.250	0.889
Concat	0.620/0.592	0.474	0.709	0.620/0.587	0.448	0.725	0.616/0.574	0.397	0.751
DAIMC	0.616/0.589	0.474	0.704	0.607/0.583	0.483	0.683	0.600/0.499	0.078	0.921
AWGF	0.646/0.644	0.638	0.651	0.554/0.549	0.526	0.571	0.580/0.561	0.483	0.640
PIC	0.600/0.532	0.250	0.815	0.620/0.528	0.147	0.910	0.574/0.605	0.733	0.476
CDIMC	0.607/0.456	0.196	0.716	0.593/0.524	0.815	0.233	0.534/0.461	0.482	0.440
MKKM_IK	0.603/0.627	0.724	0.529	0.597/0.620	0.716	0.524	0.590/0.589	0.586	0.593
Proposed	0.777/0.768	0.754	0.780	0.754/0.742	0.702	0.783	0.734/0.734	0.708	0.759

表7 本模型在HGG与LGG分类任务中的混淆矩阵结果

Tab.7 Confusion matrix results of the model for the classification task of HGG and LGG

Actual\\Predicted	10%		30%		50%
Actual\\Predicted	PN	PP	PN	PP	PN	PP
AN	88	28	81	35	81	35
AP	40	149	40	149	46	143

图2 原始特征(A)和潜在特征(B)样本二维散点图

Fig.2 Scatter plot visualization experiment of fused features (A) and original features (B) samples.

图3 所提方法在数据集上的收敛曲线

Fig.3 Convergence curve of the proposed method on the dataset.

参考文献 46

1	Tan AC, Ashley DM, López GY, et al. Management of glioblastoma: state of the art and future directions[J]. CA Cancer J Clin, 2020, 70(4): 299-312.
2	Louis DN, Perry A, Wesseling P, et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary[J]. Neuro Oncol, 2021, 23(8): 1231-51.
3	Banerjee S, Mitra S, Masulli F, et al. Deep radiomics for brain tumor detection and classification from multi-sequence MRI[J]. arXiv preprint arXiv: 1903. 09240, 2019.
4	Ellingson BM, Bendszus M, Boxerman J, et al. Consensus recommendations for a standardized Brain Tumor Imaging Protocol in clinical trials[J]. Neuro Oncol, 2015, 17(9): 1188-98.
5	Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data[J]. Radiology, 2016, 278(2): 563-77.
6	Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-62.
7	Gore S, Chougule T, Jagtap J, et al. A review of radiomics and deep predictive modeling in glioma characterization[J]. Acad Radiol, 2021, 28(11): 1599-621.
8	Wang XY, Wang DQ, Yao ZG, et al. Machine learning models for multiparametric glioma grading with quantitative result interpretations[J]. Front Neurosci, 2018, 12: 1046.
9	Bisdas S, Shen HC, Thust S, et al. Texture analysis- and support vector machine-assisted diffusional kurtosis imaging may allow in vivo gliomas grading and IDH-mutation status prediction: a preliminary study[J]. Sci Rep, 2018, 8(1): 6108-17.
10	Cho HH, Park H. Classification of low-grade and high-grade glioma using multi-modal image radiomics features[J]. Annu Int Conf IEEE Eng Med Biol Soc, 2017, 2017: 3081-4.
11	Tian Q, Yan LF, Zhang X, et al. Radiomics strategy for glioma grading using texture features from multiparametric MRI[J]. J Magn Reson Imaging, 2018, 48(6): 1518-28.
12	Sudre CH, Panovska-Griffiths J, Sanverdi E, et al. Machine learning assisted DSC-MRI radiomics as a tool for glioma classification by grade and mutation status[J]. BMC Med Inform Decis Mak, 2020, 20(1): 149-62.
13	Yang Y, Yan LF, Zhang X, et al. Glioma grading on conventional MR images: a deep learning study with transfer learning[J]. Front Neurosci, 2018, 12: 804.
14	Xu C, Tao D, Xu C. A Survey on Multi-view Learning[J]. Computer Science, 2013,4. arXiv.org.
15	Wen J, Zhang Z, Fei L, et al. A Survey on Incomplete Multiview Clustering[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2023, 53(2): 1136-49.
16	Ng A, Jordan MI, Weiss Y. On Spectral Clustering: analysis and an algorithm[J]. Adv Neural Inf Process Syst, 2001, 14.
17	Hu ML, Chen SC. Doubly aligned incomplete multi-view clustering[C]//Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence. July 13-19, 2018. Stockholm, Sweden. California: International Joint Conferences on Artificial Intelligence Organization, 2018: 2262-8.
18	Shao WX, He LF, Lu CT, et al. Online multi-view clustering with incomplete views[C]//2016 IEEE International Conference on Big Data (Big Data). December 5-8, 2016. Washington DC, USA. IEEE, 2016: 1012-7.
19	Zhao H, Liu H, Fu Y. Incomplete multi-modal visual data grouping[C]// Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence. New York: AAAI Press, 2016: 2392-8.
20	Wen J, Zhang Z, Xu Y, et al. Incomplete multi-view clustering via graph regularized matrix factorization[C]//European Conference on Computer Vision. Cham: Springer, 2019: 593-608.
21	Hu M, Chen S. One-pass incomplete multi-view clustering[C]//. Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence. Honolulu, Hawaii, USA: AAAI Press, 2019: 471.
22	Wen J, Zhang Z, Zhang Z, et al. Unified tensor framework for incomplete multi-view clustering and missing-view inferring[J]. Proc AAAI Conf Artif Intell, 2021, 35(11): 10273-81.
23	Zhang P, Wang SW, Hu JT, et al. Adaptive weighted graph fusion incomplete multi-view subspace clustering[J]. Sensors, 2020, 20(20): 5755.
24	Wen J, Xu Y, Liu H. Incomplete multiview spectral clustering with adaptive graph learning[J]. IEEE Trans Cybern, 2020, 50(4): 1418-29.
25	Liu XW, Zhu XZ, Li MM, et al. Efficient and effective incomplete multi-view clustering[C]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.
26	Liu XW, Li MM, Tang C, et al. Efficient and effective regularized incomplete multi-view clustering[J]. IEEE Trans Pattern Anal Mach Intell, 2021, 43(8): 2634-46.
27	Liu XW, Wang L, Yin JP, et al. Absent multiple kernel learning[J]. Proc AAAI Conf Artif Intell, 2015, 29(1): 2807-13.
28	Liu XW, Zhu XZ, Li MM, et al. Multiple kernel k-means with incomplete kernels[J]. IEEE Trans Pattern Anal Mach Intell, 2020, 42(5): 1191-204.
29	Xu C, Guan ZY, Zhao W, et al. Adversarial incomplete multi-view clustering[C]. IEEE Transactions on Cybernetics, 2019.
30	Wen J, Zhang Z, Xu Y, et al. CDIMC-net: cognitive deep incomplete multi-view clustering network[C]. Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence. 2021: 3230-6.
31	Wang QQ, Ding ZM, Tao ZQ, et al. Partial multi-view clustering via consistent GAN[C]//2018 IEEE International Conference on Data Mining (ICDM). November 17-20, 2018. Singapore. IEEE, 2018: 1290-5.
32	Li SY, Jiang Y, Zhou ZH. Partial multi-view clustering[J]. Proc AAAI Conf Artif Intell, 2014, 28(1): 1968-74.
33	van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational radiomics system to decode the radiographic phenotype[J]. Cancer Res, 2017, 77(21): e104-7.
34	Elmannai H, El-Rashidy N, Mashal I, et al. Polycystic ovary syndrome detection machine learning model based on optimized feature selection and explainable artificial intelligence[J]. Diagnostics, 2023, 13(8): 1506.
35	Wang H, Zong LL, Liu B, et al. Spectral perturbation meets incomplete multi-view data[J/OL]. Computer Science, 2019：.
36	Maaten L, Hinton GE. Visualizing Data using t-SNE[J]. J Mach Learn Res, 2008, 9(11).
37	Venkatesh B, Anuradha J. A review of feature selection and its methods[J]. Cybern Inf Technol, 2019, 19(1): 3-26.
38	Zebari R, Abdulazeez A, Zeebaree D, et al. A comprehensive review of dimensionality reduction techniques for feature selection and feature extraction[J]. J Appl Sci Technol Trends, 2020, 1(1): 56-70.
39	Blackwell M, Honaker J, King G. A unified approach to measurement error and missing data: overview and applications[J]. Sociol Meth Res, 2017, 46(3): 303-41.
40	Emmanuel T, Maupong T, Mpoeleng D, et al. A survey on missing data in machine learning[J]. J Big Data, 2021, 8(1): 140.
41	Zhou T, Liu MX, Thung KH, et al. Latent representation learning for Alzheimer's disease diagnosis with incomplete multi-modality neuroimaging and genetic data[J]. IEEE Trans Med Imaging, 2019, 38(10): 2411-22.
42	Kopf A, Claassen M. Latent representation learning in biology and translational medicine[J]. Patterns, 2021, 2(3): 100198.
43	Ning ZY, Xiao Q, Feng QJ, et al. Relation-induced multi-modal shared representation learning for Alzheimer's disease diagnosis[J]. IEEE Trans Med Imaging, 2021, 40(6): 1632-45.
44	Shen DG, Wu GR, Suk HI. Deep learning in medical image analysis[J]. Annu Rev Biomed Eng, 2017, 19: 221-48.
45	Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks[J]. Nature, 2017, 542(7639): 115-8.
46	Wang GT, Li WQ, Zuluaga MA, et al. Interactive medical image segmentation using deep learning with image-specific fine tuning[J]. IEEE Trans Med Imaging, 2018, 37(7): 1562-73.

[1]	张振阳, 谢金城, 钟伟雄, 梁芳蓉, 杨蕊梦, 甄鑫. 基于距匹配及判别表征学习的多模态特征融合分类模型研究：高级别胶质瘤与单发性脑转移瘤的鉴别诊断[J]. 南方医科大学学报, 2024, 44(1): 138-145.
[2]	卢明君, 屈耀铭, 马安东, 朱建彬, 邹霞, 林耕耘, 李榆欣, 刘昕孜, 温志波. 多模态MRI影像组学可预测弥漫性较低级别胶质瘤的1p/19q共缺失状态[J]. 南方医科大学学报, 2023, 43(6): 1023-1028.

基于序列缺失的MRI多序列特征填补与融合互助模型：鉴别高低级别胶质瘤

An MRI multi-sequence feature imputation and fusion mutual-aid model based on sequence deletion for differentiation of high-grade from low-grade glioma

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 46

相关文章 2

编辑推荐

Metrics

本文评价