{
  "abstract": "Early diagnosis of Alzheimer’s disease (AD) is critical for timely intervention and clinical monitoring. Electroencephalography (EEG) offers a non-invasive, low-cost, and portable alternative to neuroimaging. However, existing traditional and deep learning (DL)-based methods often face challenges related to computational cost, representational complexity, and practical deployment. In this work, we propose Fused-MSCA-2B, a two-branch multi-scale channel-attentive 1D convolutional neural network (CNN) for EEG-based AD detection. The model processes broadband multichannel EEG directly as temporal sequences and combines multi-scale temporal convolution, residual learning, squeeze-and-excitation channel attention, and feature-level branch fusion. Specifically, Branch A uses MSCA blocks with kernel configurations $$(3,5,9)$$ and $$(3,5,7)$$ to capture local and medium-range temporal dependencies, whereas Branch B uses $$(5,9,15)$$ and $$(3,5,9)$$ to capture broader and higher-level temporal patterns. The branch-level embeddings are fused and passed through dense classification layers to generate AD/HC posterior probabilities. A key component of the proposed model is an explainable AI-guided reduced-channel selection strategy. Integrated Gradients (IG) is used to rank EEG channel relevance from validation data, enabling the EEG montage to be reduced from 19 to 12 channels. We evaluated the proposed model using two benchmark datasets. The IG-guided Top-12 configuration achieved 98.60% ± 1.18% recording-level accuracy while reducing the EEG montage from 19 to 12 channels. As a supporting ablation analysis, the Fixed Top-12 configuration achieved 98.40% ± 0.73% accuracy, further confirming the effectiveness of the IG-guided reduced-channel selection strategy. The results demonstrate that the proposed XAI-guided Fused-MSCA-2B framework supports robust EEG-based AD detection with a compact and practical electrode configuration, showing strong performance with reduced channel sets.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": null,
  "author_total": 4,
  "authors": [
    {
      "name": "Jungpil Shin",
      "orcid": "0000-0002-7476-2468"
    },
    {
      "name": "Najmul Hassan",
      "orcid": "0009-0000-6499-1825"
    },
    {
      "name": "Abu Saleh Musa Miah",
      "orcid": "0000-0002-1238-0464"
    },
    {
      "name": "Satoshi Nishimura"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/806qkkre0n40k.bib",
  "cited_by": 0,
  "doi": "10.1038/s41598-026-76382-3",
  "fulltext_url": null,
  "id": "806qkkre0n40k",
  "issns": [
    "2045-2322"
  ],
  "landing_url": "https://www.nature.com/articles/s41598-026-76382-3",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0",
  "metadata_licence": null,
  "native_subjects": [
    [
      "openalex",
      "2805|EEG and Brain-Computer Interfaces"
    ]
  ],
  "openalex": "W7223016340",
  "pdf_url": "https://www.nature.com/articles/s41598-026-76382-3_reference.pdf",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-20",
  "publisher": "Springer Science and Business Media LLC",
  "ris": "https://research.draconitic.com/paper/806qkkre0n40k.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:17:31Z",
      "native_id": "10.1038/s41598-026-76382-3",
      "source": "crossref",
      "url": "https://www.nature.com/articles/s41598-026-76382-3"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "openalex:2805|EEG and Brain-Computer Interfaces",
      "subfield": "neuroscience"
    }
  ],
  "title": "EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with integrated gradients-guided channel selection",
  "title_text": "EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with integrated gradients-guided channel selection",
  "type": "peer-reviewed",
  "updated_at": "2026-10-30T12:17:31Z",
  "url": "https://research.draconitic.com/paper/806qkkre0n40k",
  "venue": "Scientific Reports",
  "versions": []
}