Scholarlink

Schizophrenia is a severe mental health condition that requires prompt and precise diagnosis. Traditional methods rely on deep convolutional neural networks (CNNs) and require large, pre-labeled image datasets, which are time-consuming and costly to compile. To solve this problem, we propose a novel active learning (AL) model that leverages deep reinforcement learning (DRL) to improve performance with fewer labeled samples. Unlike conventional methods that use heuristic selection independently of training, our approach dynamically integrates DRL to select unlabeled images for labeling. The model features multiple CNNs for feature extraction and fully connected layers for classification, trained on a limited set of labeled data and iteratively expanded via DRL-guided annotation. We also integrate a mutual learning-based artificial bee colony (ML-ABC) algorithm for optimized hyperparameter tuning. The algorithm modifies the selected food source by choosing the one with better fitness. This choice depends on a shared learning parameter between two agents. Tested on the COBRE and BrainGluSchi datasets, our model demonstrates robust accuracy with average F-measures of 93.092% and 91.238%, highlighting its efficacy and versatility in diagnosing schizophrenia.

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

Schizophrenia Detecting Using Reinforcement Learning-Based Active Learning and Hyperparameter Optimization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mohammad Mahdi Motevalli,
  • Seyed Vahid Moravvej,
  • Roohallah Alizadehsani,
  • Juan M. Gorriz

摘要

Schizophrenia is a severe mental health condition that requires prompt and precise diagnosis. Traditional methods rely on deep convolutional neural networks (CNNs) and require large, pre-labeled image datasets, which are time-consuming and costly to compile. To solve this problem, we propose a novel active learning (AL) model that leverages deep reinforcement learning (DRL) to improve performance with fewer labeled samples. Unlike conventional methods that use heuristic selection independently of training, our approach dynamically integrates DRL to select unlabeled images for labeling. The model features multiple CNNs for feature extraction and fully connected layers for classification, trained on a limited set of labeled data and iteratively expanded via DRL-guided annotation. We also integrate a mutual learning-based artificial bee colony (ML-ABC) algorithm for optimized hyperparameter tuning. The algorithm modifies the selected food source by choosing the one with better fitness. This choice depends on a shared learning parameter between two agents. Tested on the COBRE and BrainGluSchi datasets, our model demonstrates robust accuracy with average F-measures of 93.092% and 91.238%, highlighting its efficacy and versatility in diagnosing schizophrenia.