EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

A hybrid multi-stream deep learning model for cognitive feature-based schizophrenia detection

Muhilarasi Arumugam, Malarvizhi Nandagopal, Harish Kumar Shakya

Discover Computing · 06.05.2026 · 29 (1)

DOI: 10.1007/s10791-026-10136-7 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Discover Computing, online erschienen am 06.05.2026. Verfasst von Muhilarasi Arumugam und 2 weiteren Autor:innen. Die Arbeit nennt 60 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Manipal University Jaipur.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Schizophrenia is a complicated neural condition that has serious effects on thinking, behavioral and perception; thus, early and proper diagnosis is highly necessary to provide early intervention and effective treatment of the disease. Nevertheless, the current detection methods especially the classical deep learning and transfer learning paradigms have several challenges which include, overfitting on large dimensional fMRI data, poor generalizations between datasets and weaknesses in the distributions of subtle neural patterns. Such shortcomings prevent clinical dependability and relevance of automated diagnostic means on actual environments. To address these issues, this work proposes a new framework of integrating Twisted Cascade Mind Network (TwiscaMind-Net) with the Hyper-Intelligent DRaco-Hippo Algorithm (HIDRA) to be robust in schizophrenia classification. TwiscaMind-Net is a new attention-based and multi-stream convolutional neural network based on the idea of twisted processing of the brain to learn complex spatie temporal patterns in the fMRI data. The HIDRA hyper-intelligent, hippocampal neuro-like modeled and Draco-lizard survival based optimization, is an intelligence optimizer that can adjust network settings and select the most discriminating features with a bio-inspired evolutionary algorithm. The offered methodology presupposes fMRI data preprocessing on three classic datasets, including Kaggle Schizophrenia, COBRE, and FBIRN, and then deep extraction of features by using TwiscaMind-Net and optimal adjustment of parameters on the basis of HIDRA. The last clustering is done by a neural final classification layer which is fully connected. The accuracy, sensitivity, and specificity of the model on the COBRE dataset reach the remarkable value of 98.9%, 98.85%, and 98.89, respectively, with a low error rate of only 5.4 and training time of 75 s. Similar results are also on the Kaggle and the FBIRN dataset. On the whole, the TwiscaMind-Net and HIDRA framework can serve as the viable example of an effective and innovative method of schizophrenia detection that would achieve higher diagnostic accuracy, decreased computational expenses, and increased generalizability, which positions the method as an effective tool of clinical decision support in the context of mental health diagnostics.

Förderung

  • Manipal University Jaipur

Zitieren

APA

Arumugam, M., Nandagopal, M., & Shakya, H. K. (2026). A hybrid multi-stream deep learning model for cognitive feature-based schizophrenia detection. Discover Computing, 29(1). https://doi.org/10.1007/s10791-026-10136-7

Vancouver

Arumugam M, Nandagopal M, Shakya HK. A hybrid multi-stream deep learning model for cognitive feature-based schizophrenia detection. Discover Computing. 2026;29(1). doi:10.1007/s10791-026-10136-7

BibTeX
@article{Arumugam2026,
  author = {Arumugam, Muhilarasi and Nandagopal, Malarvizhi and Shakya, Harish Kumar},
  title = {A hybrid multi-stream deep learning model for cognitive feature-based schizophrenia detection},
  journal = {Discover Computing},
  year = {2026},
  volume = {29},
  number = {1},
  doi = {10.1007/s10791-026-10136-7}
}
Kostenlos starten →