Scholarlink

<p>This study addresses the challenge of improving medical image classification accuracy in the presence of class imbalance, where some categories have significantly fewer samples. A structured, multi-phase approach is proposed. First, image preprocessing is performed to ensure consistent quality and eliminate artifacts. Various segmentation models, including U-Net, FCN, SegNet, and Autoencoders are evaluated, with U-Net selected for its superior performance. Post-segmentation, multiple ensemble classifiers combining EfficientNet (B0–B3), InceptionV3, Xception, and ResNet50 are tested, and the most efficient ensemble with minimal parameters is selected. To counter class imbalance, random sampling with replacement is applied, with different ratios assessed to optimize training. Additionally, Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) are compared to fine-tune the ensemble model’s parameters. This comprehensive analysis identifies the most effective combination of preprocessing techniques, segmentation models, classification ensembles, and optimization strategies. The ultimate goal is to achieve high classification accuracy, even with imbalanced datasets, and deploy the optimized model on edge devices for real-time melanoma skin cancer prediction.</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 novel particle swarm optimization-based ensemble efficientnet learning model for imbalanced image classification in medical diagnosis: case study of melanoma skin cancer prediction

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • J. Vijaya,
  • Abhinav Roy,
  • Bhavesh Gyanchandani,
  • Jyoti Sahu

摘要

This study addresses the challenge of improving medical image classification accuracy in the presence of class imbalance, where some categories have significantly fewer samples. A structured, multi-phase approach is proposed. First, image preprocessing is performed to ensure consistent quality and eliminate artifacts. Various segmentation models, including U-Net, FCN, SegNet, and Autoencoders are evaluated, with U-Net selected for its superior performance. Post-segmentation, multiple ensemble classifiers combining EfficientNet (B0–B3), InceptionV3, Xception, and ResNet50 are tested, and the most efficient ensemble with minimal parameters is selected. To counter class imbalance, random sampling with replacement is applied, with different ratios assessed to optimize training. Additionally, Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) are compared to fine-tune the ensemble model’s parameters. This comprehensive analysis identifies the most effective combination of preprocessing techniques, segmentation models, classification ensembles, and optimization strategies. The ultimate goal is to achieve high classification accuracy, even with imbalanced datasets, and deploy the optimized model on edge devices for real-time melanoma skin cancer prediction.