CT Image Quality Classification Based on Stacking Ensemble Learning

doi:10.12068/j.issn.1005-3026.2025.20240225

Abstract

Abstract:

Computed tomography （CT） imaging has become an important form of medical imaging due to its low cost and high efficiency. However， the decline in image quality causes serious interference to diagnosis and prognosis. The limited performance of a single classifier cannot meet the requirements of high-precision CT image quality classification. To address this issue， a method based on Stacking ensemble learning was designed for artifact recognition. Based on classification diversity and individual classifier performance， random forest （RF）， back propagation neural network （BPNN）， and Inception v3， all of which are heterogeneous， were selected as the base classifiers. Extreme gradient boosting （XGBoost） was used as the meta-learner. The experimental results show that the accuracy of this method reaches 99.2%， which ensures the classification effect of the model and can meet the high accuracy requirements for CT image quality classification under the condition of an unbalanced dataset.

Key words: CT image, quality classification, Stacking ensemble learning, unbalanced dataset, complex network

CLC Number:

TP 393

Yi-wen LIU, Tao WEN, Yuan-guo BI, Hong-bo ZHU. CT Image Quality Classification Based on Stacking Ensemble Learning[J]. Journal of Northeastern University(Natural Science), 2025, 46(11): 30-36.

Figures/Tables 16

Table 1 Related symbols and their definitions

符号	定义
G	无向图
V	图G中所有节点的集合
E	图G中所有边的集合
k_m	平均度
C	平均聚类系数

Fig.1 Typical examples of artifact-absent and artifact-present images

Fig.2 Topological structure of artifact-absent and artifact-present images

Fig.3 Schematic diagram of two-phase Stacking model

Table 2 Information of dataset

数据集	总样本	正例	负例
训练集	8 546	1 763	6 783
验证集	950	196	754
测试集	1 056	218	838

Table 3 Comparison results of base classifier

模型	准确率	灵敏度	特异性
RF	93.2	85.8	95.1
XGBoost	94.1	88.5	95.6
BPNN	94.7	89.4	96.1
Inception v3	97.0	92.2	98.2

Table 4 Binary classification experiment results of

模型	准确率	灵敏度	特异性
Stacking集成模型	99.2	98.6	99.4
BPNN+Inception v3	98.1	95.9	98.7
Inception v3+RF+BPNN	98.8	97.2	99.2

Fig.4 Experiment results of accuracy comparison

Fig.5 Experiment results of sensitivity comparison

Fig.6 Experiment results of specificity comparison

Fig.7 Visualized classification results (artifact- present images)

Fig.8 Visualized classification results (artifact-absent images)

Fig.9 Boxplot of average degree

Fig.10 Boxplot of edge numbers

Fig.11 Boxplot of average clustering coefficient

Table 5 Multi-classification experiment results of

伪影程度	准确率	灵敏度
严重伪影	95.5	94.8
轻微伪影	94.8	94.4
无伪影	96.8	96.3

References 18

[1]	Ganaie M A， Hu M H， Malik A K， et al. Ensemble deep learning： a review［J］. Engineering Applications of Artificial Intelligence， 2022，115：105151.
[2]	Yu Z W， Wang D X， Zhao Z X， et al. Hybrid incremental ensemble learning for noisy real-world data classification［J］. IEEE Transactions on Cybernetics， 2019， 49（2）： 403-416.
[3]	Dang T， Nguyen T T， McCall J，et al. Two-layer ensemble of deep learning models for medical image segmentation［J］. Cognitive Computation，2024，16（3）： 1141-1160.
[4]	Ghamya K， Reddy M K. Internet of medical things-based ECG monitoring for arrhythmia classification utilizing metaheuristic optimization with ensemble deep learning model［J］. Cluster Computing， 2025，28（6）： 369.
[5]	Ganaie M A， Tanveer M. Ensemble deep random vector functional link network using privileged information for Alzheimer’s disease diagnosis［J］. IEEE/ACM Transactions on Computational Biology and Bioinformatics， 2024， 21（4）： 534-545.
[6]	Yang Y， Hu Y， Zhang X， et al. Two-stage selective ensemble of CNN via deep tree training for medical image classification［J］. IEEE Transactions on Cybernetics， 2022， 52（9）： 9194-9207.
[7]	Mohebbian M R， Walia E， Habibullah M， et al. Classifying MRI motion severity using a stacked ensemble approach［J］.Magnetic Resonance Imaging， 2021，75： 107-115.
[8]	Huertas-Tato J， Martín A， Fierrez J，et al. Fusing CNNs and statistical indicators to improve image classification［J］.Information Fusion，2022，79：174-187.
[9]	Ding J Q， Song J， Li J W， et al. Two-stage deep neural network via ensemble learning for melanoma classification［J］. Frontiers in Bioengineering and Biotechnology， 2021， 9： 758495.
[10]	Schapire R E. The strength of weak learnability［J］. Machine Learning， 1990， 5（2）： 197-227.
[11]	Boccaletti S， De Lellis P， del Genio C I， et al. The structure and dynamics of networks with higher order interactions［J］. Physics Reports， 2023， 1018： 1-64.
[12]	Yang K W， Li J C， Liu M D，et al. Complex systems and network science：a survey［J］. Journal of Systems Engineering and Electronics， 2023， 34（3）： 543-573.
[13]	Battiston F， Amico E， Barrat A， et al. The physics of higher-order interactions in complex systems［J］. Nature Physics， 2021， 17（10）： 1093-1098.
[14]	Baggio G， Bassett D S， Pasqualetti F. Data-driven control of complex networks［J］. Nature Communications， 2021， 12： 1429.
[15]	Girvan M， Newman M E J. Community structure in social and biological networks［J］. Proceedings of the National Academy of Sciences of the United States of America， 2002， 99（12）： 7821-7826.
[16]	Barabási A L， Albert R. Emergence of scaling in random networks［J］. Science， 1999， 286（5439）： 509-512.
[17]	Albert R， Barabási A L. Statistical mechanics of complex networks［J］. Reviews of Modern Physics， 2002， 74（1）： 47-97.
[18]	Hall-Beyer M. Practical guidelines for choosing GLCM textures to use in landscape classification tasks over a range of moderate spatial scales［J］. International Journal of Remote Sensing， 2017， 38（5）： 1312-1338.