轻量化AD-Net模型用于颅内肿瘤MRI图像的分类研究

doi:10.11938/cjmr20253158

摘要/Abstract

摘要：

颅内肿瘤是一种严重的神经系统疾病，早期检测对提高患者生存率具有重要意义．然而，现有深度学习模型在颅内肿瘤图像分类任务中仍面临特征提取不足、模型复杂度较高以及类别不均衡等问题．为此，本研究提出了一种轻量化深度学习网络即自适应动态网络（AD-Net）．该网络创新性地引入动态卷积机制，自适应调整滤波器响应，从而增强了对颅内肿瘤复杂、不均特征的表征能力；结合通道注意力机制，有效聚焦关键通道信息，进一步提升了分类的准确性与模型的可解释性．此外，本研究提出了结合二分类与三分类的训练策略，显著缩短了模型训练时间，降低了对计算资源的需求，使其更适用于资源受限的医疗环境．在实验中，AD-Net在准确率、精确率、召回率、F1分数以及Kappa系数等关键评价指标上均优于现有主流深度学习模型，验证了其在颅内肿瘤分类任务中的有效性与实际应用价值．

关键词: 颅内肿瘤分类, 卷积神经网络, 动态卷积, 通道注意力机制, 轻量化模型

Abstract:

Intracranial tumors represent a serious neurological disorder, and early detection is critical for improving patient survival rates. However, current deep learning models for intracranial tumor image classification often suffer from insufficient feature extraction, high model complexity, and class imbalance. To address these challenges, this study proposes a lightweight deep learning architecture, the adaptive dynamic network (AD-Net). The network innovatively incorporates a dynamic convolution mechanism that adaptively adjusts filter responses, thereby enhancing the representation of complex and imbalanced tumor features. Additionally, the integration of a channel attention mechanism enables the model to focus on critical channel information, further improving classification accuracy and interpretability. This study also introduces a combined binary and ternary classification training strategy, which significantly reduces training time and computational resource requirements, making the model more suitable for resource-constrained medical settings. Experimental results demonstrate that AD-Net outperforms existing mainstream deep learning models in accuracy, precision, recall, F1 score, and Cohen’s Kappa coefficient, confirming its effectiveness and practical value for intracranial tumor classification.

Key words: intracranial tumor classification, convolutional neural network (CNN), dynamic convolution, channel attention mechanism, lightweight model

中图分类号:

O482.53

向朝, 随力, 张昊天, 段梦雨, 刘卓睿. 轻量化AD-Net模型用于颅内肿瘤MRI图像的分类研究[J]. 波谱学杂志, 2026, 43(1): 71-86.

XIANG Zhao, SUI Li, ZHANG Haotian, DUAN Mengyu, LIU Zhuorui. A Lightweight AD-Net Model for the Classification of Intracranial Tumors in MRI Images[J]. Chinese Journal of Magnetic Resonance, 2026, 43(1): 71-86.

图/表 12

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

表1

表2

参考文献 36

[1]	AFSHAR P, MOHAMMADI A, PLATANIOTIS K N. Bayescap: A bayesian approach to brain tumor classification using capsule networks[J]. IEEE Signal Process Lett, 2020, 27: 2024-2028. doi: 10.1109/LSP.97
[2]	XUE P Y, GENG C, LI Y X, et al. A classification method for cerebral aneurysms in TOF-MRA based on improved 3D ResNet50 model[J]. Chinese J Magn Reson, 2025, 42(1): 56-66.
	薛培阳, 耿辰, 李郁欣, 等. 基于3D ResNet50改进模型的TOF-MRA脑动脉瘤分类方法[J]. 波谱学杂志, 2025, 42(1): 56-66. doi: 10.11938/cjmr20243119
[3]	PAUL J S, PLASSARD A J, LANDMAN B A, et al. Deep learning for brain tumor classification[C]// Medical Imaging 2017: Biomedical Applications in Molecular, Structural, and Functional Imaging. Orlando, United States: SPIE, 2017, 10137: 253-268.
[4]	IQBAL S, KHAN T M, NAQVI S S, et al. Ldmres-Net: a lightweight neural network for efficient medical image segmentation on iot and edge devices[J]. IEEE J Biomed Health Inform, 2023, 27(4): 1234-1242.
[5]	DAI J L, HE C, WU J, et al. A segmentation network for pancreatic cystic tumors integrating dual decoder and global attention upsampling module[J]. Chinese J Magn Reson, 2024, 41(2): 151-161.
	戴俊龙, 何聪, 武杰, 等. 融合双解码和全局注意力上采样模块的胰腺囊性肿瘤分割网络[J]. 波谱学杂志, 2024, 41(2): 151-161. doi: 10.11938/cjmr20233073
[6]	CHANG B, SUN H Y, GAO Q Y, et al. Research progress on segmentation of multi-modal cardiac medical images using traditional methods and deep learning[J]. Chinese J Magn Reson, 2024, 41(2): 224-244.
	常博, 孙灏芸, 高清宇, 等. 传统方法和深度学习用于不同模态心脏医学图像的分割研究进展[J]. 波谱学杂志, 2024, 41(2): 224-244. doi: 10.11938/cjmr20233086
[7]	AFSHAR P, PLATANIOTIS K N, MOHAMMADI A. Capsule networks for brain tumor classification based on MRI images and coarse tumor boundaries[C]// ICASSP 2019-2019 IEEE international conference on acoustics, speech and signal processing (ICASSP). Athens, Greece: IEEE, 2019: 1368-1372.
[8]	FENG L, WU K, PEI Z, et al. MLU-Net: A multi-level lightweight U-Net for medical image segmentation integrating frequency representation and MLP-based methods[J]. IEEE Access, 2024, 12: 20734-20751. doi: 10.1109/ACCESS.2024.3360889
[9]	XU Z S, YUAN X H, HUANG Z H, et al. Multi-source feature classification model for pancreatic mucinous and serous cystic neoplasms based on deep learning[J]. Chinese J Magn Reson, 2024, 41(1): 19-29.
	徐真顺, 袁小涵, 黄子珩, 等. 基于深度学习的胰腺黏液性和浆液性囊性肿瘤的多源特征分类模型[J]. 波谱学杂志, 2024, 41(1): 19-29. doi: 10.11938/cjmr20233064
[10]	SWATI Z N K, ZHAO Q, KABIR M, et al. Brain tumor classification for MR images using transfer learning and fine-tuning[J]. Comput Med Imaging Graph, 2019, 75: 34-46. doi: 10.1016/j.compmedimag.2019.05.001
[11]	GHASSEMI N, SHOEIBI A, ROUHANI M. Deep neural network with generative adversarial networks pre-training for brain tumor classification based on MR images[J]. Biomed Signal Process Control, 2020, 57: 101678. doi: 10.1016/j.bspc.2019.101678
[12]	AFSHAR P, PLATANIOTIS K N, MOHAMMADI A. BoostCaps:a boosted capsule network for brain tumor classification[C]// 2020 42nd annual international conference of the IEEE engineering in medicine & biology society (EMBC). Montreal, QC, Canada: IEEE, 2020: 1075-1079.
[13]	BODAPATI J D, SHAIK N S, NARALASETTI V, et al. Joint training of two-channel deep neural network for brain tumor classification[J]. Signal Image Video P, 2021, 15(4): 753-760.
[14]	KALKHOF J, GONZÁLEZ C, MUKHOPADHYAY A. Med-nca: Robust and lightweight segmentation with neural cellular automata[C]// International Conference on Information Processing in Medical Imaging. San Antonio, Texas, USA. Cham: Springer Nature Switzerland, 2023: 705-716.
[15]	ABIRAMI S, VENKATESAN G K D P. Deep learning and spark architecture based intelligent brain tumor MRI image severity classification[J]. Biomed Signal Process Control, 2022, 76: 103644. doi: 10.1016/j.bspc.2022.103644
[16]	WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C]// Proceedings of the European conference on computer vision (ECCV). Munich, Germany. Cham: Springer, 2018: 3-19.
[17]	HAN B, XU J, WANG Y J, et al. Triple classification of BI-RADS 3-5 breast lesions based on MRI radiomics[J]. Chinese J Magn Reson, 2023, 40(1): 52-67.
	韩冰, 徐晶, 王远军, 等. 基于MRI影像组学的BI-RADS 3-5类乳腺病变三分类[J]. 波谱学杂志, 2023, 40(1): 52-67. doi: 10.11938/cjmr20222971
[18]	ZHANG H, GOODFELLOW I, METAXAS D, et al. Self-attention generative adversarial networks[C]// International conference on machine learning. Long Beach, CA, USA. PMLR, 2019: 7354-7363.
[19]	CAO Y, XU J, LIN S, et al. Gcnet: Non-local networks meet squeeze-excitation networks and beyond[C]// Proceedings of the IEEE/CVF international conference on computer vision workshops. Seoul, South Korea: IEEE, 2019: 1971-1980.
[20]	SHARMA P, NAYAK D R, BALABANTARAY B K, et al. A survey on cancer detection via convolutional neural networks: Current challenges and future directions[J]. Neural Netw, 2024, 169: 637-659. doi: 10.1016/j.neunet.2023.11.006
[21]	BODAPATI J D, SHAREEF S N, NARALASETTI V, et al. Msenet: Multi-modal squeeze-and-excitation network for brain tumor severity prediction[J]. Int J Pattern Recognit Artif Intell, 2021, 35(7): 2157005. doi: 10.1142/S0218001421570056
[22]	DUTTA T K, NAYAK D R. CDANet:Channel split dual attention based CNN for brain tumor classification in MR images[C]// 2022 IEEE international conference on image processing (ICIP). Bordeaux, France: IEEE, 2022: 4208-4212.
[23]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. Las Vegas, NV, USA: IEEE, 2016: 770-778.
[24]	HASSANI A, WALTON S, SHAH N, et al. Escaping the big data paradigm with compact transformers[J]. arXiv: 2104.05704, 2021.
[25]	LI K, WANG Y, ZHANG J, et al. Uniformer: Unifying convolution and self-attention for visual recognition[J]. IEEE Trans Pattern Anal Mach Intell, 2023, 45(10): 12581-12600. doi: 10.1109/TPAMI.2023.3282631
[26]	ALANAZI M F, ALI M U, HUSSAIN S J, et al. Brain tumor/mass classification framework using magnetic-resonance-imaging-based isolated and developed transfer deep-learning model[J]. Sensors, 2022, 22(1): 372. doi: 10.3390/s22010372
[27]	ZAHOOR M M, QURESHI S A, BIBI S, et al. A new deep hybrid boosted and ensemble learning-based brain tumor analysis using MRI[J]. Sensors, 2022, 22(7): 2726. doi: 10.3390/s22072726
[28]	CHEN Y, DAI X, LIU M, et al. Dynamic convolution: Attention over convolution kernels[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. Seattle, Washington, USA: IEEE, 2020: 11030-11039.
[29]	CAO R, YANG F, MA S C, et al. Development and interpretation of a pathomics-based model for the prediction of microsatellite instability in Colorectal Cancer[J]. Theranostics, 2020, 10(24): 11080. doi: 10.7150/thno.49864 pmid: 33042271
[30]	NAWAZ K, ZANIB A, SHABIR I, et al. Skin cancer detection using dermoscopic images with convolutional neural network[J]. Sci Rep, 2025, 15(1): 7252. doi: 10.1038/s41598-025-91446-6
[31]	MAMUN O R, RASHID M D, NASHBAT M, ASHRAF A, et al. Self-DSNet: A novel self-ONNs based deep learning framework for multimodal driving distraction detection[J]. IEEE Access, 2025, 13: 42322-42335. doi: 10.1109/ACCESS.2025.3545359
[32]	YIN S, LI H, TENG L, et al. Brain CT image classification based on mask RCNN and attention mechanism[J]. Sci Rep, 2024, 14(1): 29300. doi: 10.1038/s41598-024-78566-1
[33]	ZENG L, ZHANG H H. Robust brain MRI image classification with SIBOW-SVM[J]. Comput Med Imaging Graph, 2024, 118: 102451. doi: 10.1016/j.compmedimag.2024.102451
[34]	HUA C, CHEN Y, TAO J, et al. Dual-pathway EEG model with channel attention for virtual reality motion sickness detection[J]. J Neurosci Methods, 2025, 418: 110425. doi: 10.1016/j.jneumeth.2025.110425
[35]	DUAN X, CHEN Y, DUAN X, et al. Improved grain boundary reconstruction method based on channel attention mechanism[J]. Materials, 2025, 18(2): 253. doi: 10.3390/ma18020253
[36]	AL-KHULAIDI N, ALMOURISH M H, TALEB E M A, et al. Improving detection and classification of brain tumors using DenseNet201[J]. J Sci Technol, 2025, 30(6): 2855.

模型名称	准确率(Accuracy)	精确率(Precision)	召回率(Recall)	特异性(Specifity)	F1分数(F1 Score)
Self-ONN4L1DN	0.9390±0.0129	0.9348±0.0133	0.9377±0.013	0.9434±0.0125	0.9350±0.0133
Self-ONN4L	0.9376±0.0130	0.9356±0.0132	0.9376±0.0130	0.9428±0.0125	0.9375±0.0131
Self-ONN6L	0.9419±0.0126	0.9485±0.0119	0.9429±0.0125	0.9547±0.0112	0.9458±0.0122
Self-ONN6L1DN	0.9650±0.0112	0.9553±0.0111	0.9520±0.0115	0.9611±0.0104	0.9531±0.0114
Vanilla CNN8L	0.9295±0.0138	0.9289±0.0139	0.9292±0.0138	0.9395±0.0129	0.9278±0.0140
Vanilla CNN6L	0.9283±0.0139	0.9276±0.0140	0.9290±0.0139	0.9387±0.0129	0.9245±0.0143
DenseNet201	0.9458±0.0122	0.9455±0.0122	0.9428±0.0125	0.9584±0.0108	0.9480±0.0120
ResNet101	0.9490±0.0107	0.9596±0.0106	0.9589±0.0107	0.9589±0.0107	0.9586±0.0107
AD-Net	0.9747±0.0092	0.9693±0.0093	0.9685±0.0094	0.9795±0.0076	0.9683±0.0095

模型名称	准确率(Accracy)	精确率(Precision)	召回率(Recall)	F1分数(F1 Score)	Kappa
ResNet50	0.9362	0.9251	0.9331	0.9434	0.9325
ResNet50+Channel Attention	0.9519	0.9369	0.9550	0.9552	0.9484
ResNet50+Dynamic Convolution	0.9601	0.9553	0.9433	0.9565	0.9526
AD-Net	0.9747	0.9693	0.9685	0.9683	0.9773