Scholarlink

In this comprehensive work on PCOD classification using ensemble models, we consider the ultrasound images of patients diagnosed with and without PCOD. Out of the total 5167 images comprising the dataset, 2567 represent PCOD and the rest 2600 represent non-PCOD patients, to prevent class imbalance. Following the creation of the dataset is the pre-processing using 2D filters, Gaussian Blur, and Mean and Median Filters which are used to enhance the images as seen in the results section. Following this step is the implementation of segmentation algorithms whose performance is measured using IoU and mAP. The various algorithms used to perform segmentation are EfficientNet, Hybrid CNN, Transformers, and Attention Nets. The proposed ensemble model which comprises Transformers and Attention Nets outperforms the individual models by providing an accuracy of 91.07% with an F1 score of 88.89% for PCOS and an accuracy of 93.46% with an F1 score of 93.09% for non-PCOD conditions. The results of this study show that the proposed ensemble technique can efficiently aid in diagnosing PCOD using ultrasound images, which may develop into creating an effective computer-based PCOD diagnosis system.

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 Imaging: A Holistic Exploration of Precise PCOD Segmentation and Classification Using Ensemble Deep Learning Approaches

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ashwini Kodipalli,
  • Trupthi Rao,
  • Taha Ismail,
  • Neha Nayak

摘要

In this comprehensive work on PCOD classification using ensemble models, we consider the ultrasound images of patients diagnosed with and without PCOD. Out of the total 5167 images comprising the dataset, 2567 represent PCOD and the rest 2600 represent non-PCOD patients, to prevent class imbalance. Following the creation of the dataset is the pre-processing using 2D filters, Gaussian Blur, and Mean and Median Filters which are used to enhance the images as seen in the results section. Following this step is the implementation of segmentation algorithms whose performance is measured using IoU and mAP. The various algorithms used to perform segmentation are EfficientNet, Hybrid CNN, Transformers, and Attention Nets. The proposed ensemble model which comprises Transformers and Attention Nets outperforms the individual models by providing an accuracy of 91.07% with an F1 score of 88.89% for PCOS and an accuracy of 93.46% with an F1 score of 93.09% for non-PCOD conditions. The results of this study show that the proposed ensemble technique can efficiently aid in diagnosing PCOD using ultrasound images, which may develop into creating an effective computer-based PCOD diagnosis system.