Scholarlink

Purpose <p>Understanding visual stimuli is becoming an increasingly challenging task in neuroscience. The objective is to categorize the brain activity patterns that occur when individuals view visual objects. This study concentrates on classifying images, particularly looking at how visual representations are decoded using functional magnetic resonance imaging (fMRI) with an ensemble approach.</p> Methods <p>The investigated databases were sourced from the “openneuro.org” online platforms: DS105, DS107, and DS116. Three diverse base learners—CNN, RNN, and GRU—were employed, and metadata was generated. To manage uncertainties, a fuzzy min–max (FMM) model was incorporated for detection. Regarding class probability and labels, the fuzzy scheme received the ensemble outputs from the basic learners. The min–max method was employed by the fuzzy model to make precise judgments. The evaluation metrics used for assessment included precision, recall, accuracy, sensitivity, specificity, and F1-score.</p> Outcomes <p>The achieved validation accuracy rates for visual representation decoding were 98.91% for DS105, 97.63% for DS107, and 99.34% for DS116, as shown in the results section.</p> Conclusion <p>The proposed method achieves test accuracy rates greater than 97.50% across three different fMRI databases, making it competitive with current methodologies.</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

A new fuzzy-based deep ensemble framework for visual representation decoding from fMRI brain data

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Yan Yang

摘要

Purpose

Understanding visual stimuli is becoming an increasingly challenging task in neuroscience. The objective is to categorize the brain activity patterns that occur when individuals view visual objects. This study concentrates on classifying images, particularly looking at how visual representations are decoded using functional magnetic resonance imaging (fMRI) with an ensemble approach.

Methods

The investigated databases were sourced from the “openneuro.org” online platforms: DS105, DS107, and DS116. Three diverse base learners—CNN, RNN, and GRU—were employed, and metadata was generated. To manage uncertainties, a fuzzy min–max (FMM) model was incorporated for detection. Regarding class probability and labels, the fuzzy scheme received the ensemble outputs from the basic learners. The min–max method was employed by the fuzzy model to make precise judgments. The evaluation metrics used for assessment included precision, recall, accuracy, sensitivity, specificity, and F1-score.

Outcomes

The achieved validation accuracy rates for visual representation decoding were 98.91% for DS105, 97.63% for DS107, and 99.34% for DS116, as shown in the results section.

Conclusion

The proposed method achieves test accuracy rates greater than 97.50% across three different fMRI databases, making it competitive with current methodologies.