Scholarlink

Purpose <p>Ovarian tumors pose significant diagnostic challenges in gynecology. Accurate and early identification is critical for effective clinical management, fertility preservation, and timely tumor intervention. Ultrasound imaging is the primary diagnostic tool for ovarian evaluation due to its noninvasive, cost-effective, and widely accessible nature. However, challenges, such as poor contrast, speckle noise, and unclear image boundaries, hinder its full diagnostic potential, necessitating the development of advanced methods to improve image quality and accuracy.</p> Methods <p>This research presents a deep learning-based framework for the automated detection and classification of ovarian tumors in ultrasound images. The process begins with homomorphic filtering to preprocess ultrasound images of ovarian tumors, enhancing their details and contrast. For accurate segmentation, an enhanced SegNet model with an atrous spatial pyramid pooling module (ASPPM) is then used. The segmented images are then used to extract features that help classify ovarian tumors, such as refined Local Arc patterns. Furthermore, a novel hybrid model, known as modified LinkNet + DenseNet, is suggested for ovarian tumor detection, initiated from the feature extraction stage. Lastly, an advanced score-level fusion technique integrates the outputs of both classifiers, ensuring reliable detection outcomes.</p> Results <p>The improved LinkNet + DenseNet system obtains a detection accuracy of 0.953 at 70% training data, when compared to existing models.</p> Conclusion <p>This comprehensive approach, incorporating advanced techniques from image processing and machine learning, offers a robust solution for the accurate detection and classification of ovarian tumors, thereby supporting improved diagnostic reliability and clinical decision-making.</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

Novel Deep Learning-Based Framework for Multi-class Detection of Ovarian Tumors Using Ultrasound Imaging

  • 期刊论文

  • 发表日期：

    2025年8月17日

查看全文
  • Namani Deepika Rani,
  • Mahesh Babu Arrama

摘要

Purpose

Ovarian tumors pose significant diagnostic challenges in gynecology. Accurate and early identification is critical for effective clinical management, fertility preservation, and timely tumor intervention. Ultrasound imaging is the primary diagnostic tool for ovarian evaluation due to its noninvasive, cost-effective, and widely accessible nature. However, challenges, such as poor contrast, speckle noise, and unclear image boundaries, hinder its full diagnostic potential, necessitating the development of advanced methods to improve image quality and accuracy.

Methods

This research presents a deep learning-based framework for the automated detection and classification of ovarian tumors in ultrasound images. The process begins with homomorphic filtering to preprocess ultrasound images of ovarian tumors, enhancing their details and contrast. For accurate segmentation, an enhanced SegNet model with an atrous spatial pyramid pooling module (ASPPM) is then used. The segmented images are then used to extract features that help classify ovarian tumors, such as refined Local Arc patterns. Furthermore, a novel hybrid model, known as modified LinkNet + DenseNet, is suggested for ovarian tumor detection, initiated from the feature extraction stage. Lastly, an advanced score-level fusion technique integrates the outputs of both classifiers, ensuring reliable detection outcomes.

Results

The improved LinkNet + DenseNet system obtains a detection accuracy of 0.953 at 70% training data, when compared to existing models.

Conclusion

This comprehensive approach, incorporating advanced techniques from image processing and machine learning, offers a robust solution for the accurate detection and classification of ovarian tumors, thereby supporting improved diagnostic reliability and clinical decision-making.