Scholarlink

<p>Medical image analysis holds major significance in modern healthcare, which is essential for both diagnosing diseases and planning treatment. The development of advanced imaging mechanisms like Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) has become a necessary instrument in clinical practice. Through the analysis of these imaging techniques, medical professionals receive detailed insights into anatomic structures and pathological conditions. Yet, it poses challenges related to classification accuracy and privacy measures, demanding innovative approaches to ensure reliable diagnosis and patient confidentiality. To tackle these problems, we introduce a novel approach integrating Federated Learning with a hybrid optimization algorithm (FL Brain-HOA) to enhance brain tumor detection accuracy as well as privacy. This model enables collaborative training across multiple decentralized clients without sharing raw medical data, there by preserving patient privacy To dynamically tune the hyperparameters of the CNN models, such as the GA-SA optimizer is used and deployed locally at each client, where the model accuracy and performs well in correctly detecting and classifying brain tumors even in the non-IID situation. A fitness-score-guided aggregation strategy is introduced to enhance global model convergence by prioritizing high-quality local updates. The model further incorporates Gradient-weighted Class Activation Mapping (Grad-CAM) for visual interpretability, allowing clinicians to understand the rationale behind each prediction. The FL Brain-HOA model demonstrates strong performance in evaluating brain tumor MRI and Brats 19 datasets with 98.75% of accuracy, 98.32% of precision, 97.86% of recall, 98.08% of F1-score, and 0.987 AUC-ROC in brain tumor MRI dataset and obtained 98.02% accuracy, 98.23% precision, 97.65% recall and 97.93% of F1-score in Brats 2019 dataset. The simulation result indicates our FL Brain-HOA model is superior in the detection of brain tumors and improves categorization accuracy while supporting the principles of data privacy and paving the way for more effective diagnosis, treatment, and management of brain tumors.</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

Enhancing Medical Image Analysis with FL Brain-HOA: A Federated Learning and Hybrid Optimization Approach

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Jeenath Shafana Nagoor kani,
  • Senthilselvi Ayothi

摘要

Medical image analysis holds major significance in modern healthcare, which is essential for both diagnosing diseases and planning treatment. The development of advanced imaging mechanisms like Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) has become a necessary instrument in clinical practice. Through the analysis of these imaging techniques, medical professionals receive detailed insights into anatomic structures and pathological conditions. Yet, it poses challenges related to classification accuracy and privacy measures, demanding innovative approaches to ensure reliable diagnosis and patient confidentiality. To tackle these problems, we introduce a novel approach integrating Federated Learning with a hybrid optimization algorithm (FL Brain-HOA) to enhance brain tumor detection accuracy as well as privacy. This model enables collaborative training across multiple decentralized clients without sharing raw medical data, there by preserving patient privacy To dynamically tune the hyperparameters of the CNN models, such as the GA-SA optimizer is used and deployed locally at each client, where the model accuracy and performs well in correctly detecting and classifying brain tumors even in the non-IID situation. A fitness-score-guided aggregation strategy is introduced to enhance global model convergence by prioritizing high-quality local updates. The model further incorporates Gradient-weighted Class Activation Mapping (Grad-CAM) for visual interpretability, allowing clinicians to understand the rationale behind each prediction. The FL Brain-HOA model demonstrates strong performance in evaluating brain tumor MRI and Brats 19 datasets with 98.75% of accuracy, 98.32% of precision, 97.86% of recall, 98.08% of F1-score, and 0.987 AUC-ROC in brain tumor MRI dataset and obtained 98.02% accuracy, 98.23% precision, 97.65% recall and 97.93% of F1-score in Brats 2019 dataset. The simulation result indicates our FL Brain-HOA model is superior in the detection of brain tumors and improves categorization accuracy while supporting the principles of data privacy and paving the way for more effective diagnosis, treatment, and management of brain tumors.