Scholarlink

<p>Electroencephalography (EEG) is a non-invasive, high-temporal-resolution method for diagnosing and monitoring neurological disorders. Deep learning has recently substantially enhanced the state of the art for automated EEG analysis. However, many of the currently applied paradigms are still challenged by limited generalisation across datasets, vulnerability to noise or preprocessing changes, and the absence of interpretable decision rules. Additionally, many deep learning models operate like black boxes, which limits their use in clinical settings where interpretability and trust are key. Although the potential of quantum-inspired learning has recently been demonstrated through improved feature separability in high-dimensional signal spaces, its scope of applicability does not yet extend to deep temporal modelling and explainable artificial intelligence applications. We address these limitations by introducing QuantumNeuroXAI, a quantum-inspired deep learning framework implemented on classical hardware that leverages structured feature encoding inspired by quantum neural networks to provide inherent explainability for EEG-based diagnosis of neurological disorders. This framework hybridises quantum-inspired feature encoding with a deep-learning architecture that blends temporal convolutional and attention-based recurrent modelling to capture local and long-range patterns of dependencies in EEG signals. The framework incorporates a multi-level explainability module relevant at the signal, model, and quantum-representation levels, allowing predictions to be interpreted in a clinically meaningful and transparent fashion. We conduct extensive experiments on three publicly available EEG datasets (TUH EEG, CHB-MIT, and BCI Competition IV-2a) to evaluate the proposed framework. These quantitative results show that QuantumNeuroXAI achieves statistically significant and large effect sizes, with macro-F1 improvements of up to 5.2% over classical machine learning, deep learning, and hybrid baseline models. Additional robustness and scalability analyses further validate stable performance against dataset shift and across various preprocessing configurations. In summary, QuantumNeuroXAI is an interpretable and reproducible solution for EEG-based neurological analysis, demonstrating promise for clinical decision support and future scalability to multimodal brain signal applications. It is important to note that the proposed framework does not rely on quantum hardware and is fully implemented using classical computational resources. The implementation of the proposed framework is publicly available at: <a href="https://github.com/venkateshwarlu-bondu/QuantumNeuroXAI">https://github.com/venkateshwarlu-bondu/QuantumNeuroXAI</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

QuantumNeuroXAI: a quantum-inspired deep learning framework with explainability for brain signal analysis and neurological disorder detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月10日

查看全文
  • T. Gayathri,
  • G. Manjula,
  • Harish H. Kenchannavar,
  • Danthuluri Sudha,
  • Santosh Kumar Jankatti,
  • Ramandeep Kaur,
  • Bondu Venkateswarlu

摘要

Electroencephalography (EEG) is a non-invasive, high-temporal-resolution method for diagnosing and monitoring neurological disorders. Deep learning has recently substantially enhanced the state of the art for automated EEG analysis. However, many of the currently applied paradigms are still challenged by limited generalisation across datasets, vulnerability to noise or preprocessing changes, and the absence of interpretable decision rules. Additionally, many deep learning models operate like black boxes, which limits their use in clinical settings where interpretability and trust are key. Although the potential of quantum-inspired learning has recently been demonstrated through improved feature separability in high-dimensional signal spaces, its scope of applicability does not yet extend to deep temporal modelling and explainable artificial intelligence applications. We address these limitations by introducing QuantumNeuroXAI, a quantum-inspired deep learning framework implemented on classical hardware that leverages structured feature encoding inspired by quantum neural networks to provide inherent explainability for EEG-based diagnosis of neurological disorders. This framework hybridises quantum-inspired feature encoding with a deep-learning architecture that blends temporal convolutional and attention-based recurrent modelling to capture local and long-range patterns of dependencies in EEG signals. The framework incorporates a multi-level explainability module relevant at the signal, model, and quantum-representation levels, allowing predictions to be interpreted in a clinically meaningful and transparent fashion. We conduct extensive experiments on three publicly available EEG datasets (TUH EEG, CHB-MIT, and BCI Competition IV-2a) to evaluate the proposed framework. These quantitative results show that QuantumNeuroXAI achieves statistically significant and large effect sizes, with macro-F1 improvements of up to 5.2% over classical machine learning, deep learning, and hybrid baseline models. Additional robustness and scalability analyses further validate stable performance against dataset shift and across various preprocessing configurations. In summary, QuantumNeuroXAI is an interpretable and reproducible solution for EEG-based neurological analysis, demonstrating promise for clinical decision support and future scalability to multimodal brain signal applications. It is important to note that the proposed framework does not rely on quantum hardware and is fully implemented using classical computational resources. The implementation of the proposed framework is publicly available at: https://github.com/venkateshwarlu-bondu/QuantumNeuroXAI.