Scholarlink

Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting reproductive-aged women. Early and accurate detection of PCOS is crucial for timely intervention and treatment. In this study, we performed a comprehensive comparative analysis of six deep learning models, Custom Convolutional Neural Network, VGG16, EfficientNet V2B3, DenseNet121, ResNet50, and Inception V3, to distinguish between PCOS and non-PCOS images. Applying cross-validation to improve model evaluation and address the issue of overfitting is a novel contribution of this work, not previously documented in the literature. Our findings came up with VGG16 as the best-performed model with a validation accuracy of 99.61% and test accuracy of 100%. Further, this study is the first to report the AUC-ROC metric to provide deeper insight into model effectiveness. The study highlights the potential of other evaluation metrics for performance evaluation and paves the way for future improvements and real-world clinical applications.

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

Comparative Analysis of Deep Learning Architectures for PCOS Detection: A Cross-Validation Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kirti Sharma,
  • Bhawna Jain

摘要

Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting reproductive-aged women. Early and accurate detection of PCOS is crucial for timely intervention and treatment. In this study, we performed a comprehensive comparative analysis of six deep learning models, Custom Convolutional Neural Network, VGG16, EfficientNet V2B3, DenseNet121, ResNet50, and Inception V3, to distinguish between PCOS and non-PCOS images. Applying cross-validation to improve model evaluation and address the issue of overfitting is a novel contribution of this work, not previously documented in the literature. Our findings came up with VGG16 as the best-performed model with a validation accuracy of 99.61% and test accuracy of 100%. Further, this study is the first to report the AUC-ROC metric to provide deeper insight into model effectiveness. The study highlights the potential of other evaluation metrics for performance evaluation and paves the way for future improvements and real-world clinical applications.