| 41 | 0 | 135 |
| 下载次数 | 被引频次 | 阅读次数 |
为了解决阿尔茨海默病脑电分类中病理特征易被背景节律掩盖、注意力头易发生模式坍缩的问题，本文提出一种相位感知周期Transformer分类框架PACformer。该模型首先利用由大感受野一维卷积与多层感知机组成的自适应相位投影器估计脑电片段起始相位，并通过相位感知门控对嵌入特征进行动态调制，实现相位对齐与特征增强；随后在编码器中引入可学习傅里叶级数，对背景周期趋势进行显式建模，并采用“减去—关注—加回”的残差分解策略突出局部病理扰动；最后设计多头谱多样性损失，约束不同注意力头覆盖互补频段，抑制频谱表示坍缩。在APAVA 数据集上的实验结果表明，PACformer 的准确率、受试者工作特征曲线下面积（AUROC）和精确率-召回率曲线下面积（AUPRC）分别达到 77.75%、86.56% 和 87.02%，优于 11 种对比模型。结果表明，显式的相位对齐与频域分解有助于提升非平稳生理信号的判别建模能力，可为基于脑电的阿尔茨海默病辅助筛查提供更稳健的技术方案。
Abstract：To address the challenges of background rhythm interference and attention head mode collapse in Alzheimer’s disease electroencephalography classification， we propose a phase-aware cyclical Transformer framework， termed PACformer. The model first employs an adaptive phase projector， composed of a large-kernel 1D convolution and a multilayer perceptron， to estimate the starting phase of each EEG epoch and generate phase-aware gating weights for refining token embeddings. It then introduces a Fourier residual decomposition encoder that learns periodic background trends with a trainable Fourier series and follows a subtract-attend-add strategy to separate stationary rhythms from pathology-related residuals. In addition， a multi-head spectral diversity loss is designed to encourage different attention heads to cover complementary frequency bands. Experiments on the APAVA dataset show that PACformer achieves 77.75% accuracy， 86.56% AUROC， and 87.02% AUPRC， outperforming 11 competitive baselines. These results indicate that explicit phase alignment and frequency-domain decomposition improve the modeling of non-stationary physiological signals and support more robust EEG-based AD screening.
[1] BREIJYEH Z， KARAMAN R. Comprehensive review on Alzheimer’s disease: causes and treatment[J]. Molecules， 2020， 25(24): 5789.
[2] FERRI C P， PRINCE M， BRAYNE C， et al. Global prevalence of dementia: a Delphi consensus study[J]. The Lancet， 2005， 366(9503): 2112-2117.
[3] BLENNOW K， DE LEON M J， ZETTERBERG H. Alzheimer’s disease[J]. The Lancet， 2006， 368(9533): 387-403.
[4] AHMED S， MOMIN M， REN J， et al. Stick-and-play bioadhesive hairlike electrodes for chronic EEG recording on human[J]. npj Biomedical Innovations， 2025， 2(1): 9.
[5] CASSANI R， ESTARELLAS M， SAN-MARTIN R， et al. Systematic review on resting-state EEG for Alzheimer’s disease diagnosis and progression assessment[J]. Disease Markers， 2018: 5174815.
[6] JEONG J. EEG dynamics in patients with Alzheimer’s disease[J]. Clinical Neurophysiology， 2004， 115(7): 1490-1505.
[7] CHEN T， WANG C， ZHANG Y， et al. MFS-Fusion: Mamba-integrated deep multi-modal image fusion framework with multi-scale Fourier enhancement and spatial calibration[J]. Expert Systems with Applications， 2025: 130054.
[8] IERACITANO C， MAMMONE N， BRAMANTI A， et al. A convolutional neural network approach for classification of dementia stages based on 2D-spectral representation of EEG recordings[J]. Neurocomputing， 2019， 323: 96-107.
[9] VASWANI A， SHAZEER N， PARMAR N， et al. Attention is All You Need[C]//Advances in Neural Information Processing Systems 30， December 4-9， 2017， Long Beach， CA， USA. Cambridge: MIT Press， 2017: 5998-6008.
[10] VAFAEI E， HOSSEINI M. Transformers in EEG analysis: a review of architectures and applications in motor imagery， seizure， and emotion classification[J]. Sensors， 2025， 25(5): 1293.
[11] WANG Y， HUANG N， LI T， et al. Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification[C]//Advances in Neural Information Processing Systems 37， December 10-15， 2024， Vancouver， Canada. Cambridge: MIT Press， 2024: 36314-36341.
[12] ZHANG Y， MA L， PAL S， et al. Multi-resolution time-series transformer for long-term forecasting[C]//Proceedings of The 27th International Conference on Artificial Intelligence and Statistics， May 2-4， 2024， Palau de Congressos， Valencia， Spain. JMLR， 2024: 4222-4230.
[13] WU H， XU J， WANG J， et al. Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting[C]//Advances in Neural Information Processing Systems 34， December 6-14， 2021， virtual. Cambridge: MIT Press， 2021: 22419-22430.
[14] ESCUDERO J， ABáSOLO D， HORNERO R， et al. Analysis of electroencephalograms in Alzheimer’s disease patients with multiscale entropy[J]. Physiological Measurement， 2006， 27(11): 1091-1106.
[15] ZHANG Y， YAN J. Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting[C]//The Eleventh International Conference on Learning Representations， May 1-5， 2023， Kigali， Rwanda. 2023: 5268-5288.
[16] ZHOU T， MA Z， WEN Q， et al. FEDformer: Frequency enhanced decomposed Transformer for long-term series forecasting[C]//Proceedings of the 39th International Conference on Machine Learning， July 17-23， 2022， Baltimore， Maryland， USA. New York: ACM， 2022: 27268-27286.
[17] ZHOU H， ZHANG S， PENG J， et al. Informer: Beyond efficient Transformer for long sequence time-series forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Menlo Park， CA: AAAI， 2021， 35(12): 11106-11115.
[18] LIU Y， HU T， ZHANG H， et al. iTransformer: Inverted transformers are effective for time series forecasting[C]//The Twelfth International Conference on Learning Representations， May 7-11， 2024， Vienna， Austria. 2024: 4004-4028.
[19] LIU Y， WU H， WANG J， et al. Non-stationary Transformers: Exploring the stationarity in time series forecasting[C]//Advances in Neural Information Processing Systems 35， November 28-December 9， 2022， New Orleans， Louisiana， USA. Cambridge: MIT Press， 2022: 9881-9893.
[20] NIE Y， NGUYEN N H， SINTHONG P， et al. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers[C]//The Eleventh International Conference on Learning Representations， May 1-5， 2023， Kigali， Rwanda. 2023: 33132-33155.
[21] KITAEV N， KAISER ?， LEVSKAYA A. Reformer: The efficient transformer[C]//8th International Conference on Learning Representations， April 26-30， 2020， Addis Ababa， Ethiopia. 2020: 12764-12775.
基本信息:
中图分类号:TN911.7;R749.16;R741
引用信息:
[1]葛佳昌,蒋亦樟,黄丽军,等.基于相位感知与傅里叶分解的阿尔茨海默病脑电分类[J].南通大学学报(自然科学版)().
基金信息:
山东省自然科学基金项目（ZR2024MF134）; 江苏省“333工程”高层次人才资助项目; 江苏省卫健委老年健康资助性项目（LKM2024038）; 苏州市“科教强卫”重点项目（ZDXM2024017）; 苏州医学重点学科（卫生信息管理学）项目（SZXK202528）; 常熟市医学人工智能与大数据重点实验室项目（CYZ202301，CS202314）; 苏州市数据创新应用实验室项目; 江南大学江苏研究生工作站项目
2026-05-22
2026-05-22
2026-05-22