Scholarlink

<p>The diffuse and advanced stage of liver disease is identified as liver cirrhosis. The detection of liver illness depends on the full automatic diagnosis procedure that identifies the liver and tumor regions from the image modalities. Moreover, many recent works have utilized a single image modality to classify and segment liver cirrhosis. These modalities, however, lack greater detection accuracy and are biased. Therefore, this work applies a novel model to improve the efficacy of liver tumor analysis by merging deep learning with image modalities. In this work, multimodality images (MMI), such as CT, MRI, and US, are taken into account. First, the noise in the input MMI is removed using a Median Filter (MF) and an Extended Guided Filter (EGF). Furthermore, contrast-limited adaptive histogram equalization (CLAHE), which maintains image brightness while enhancing original contrast, enhances the quality of liver multimodality images. After that, the Advanced Gray-Level Co-occurrence Matrix (AGLCM) facilitates the Feature Extraction (FE) function. To further improve accuracy, the key features were chosen using the Lotus Effect Optimization Algorithm (LEOA). After that, the proposed Optimal Fine-Tuned Deep Neural Network (OFT-DNN)-based deep learning model is used to categorize the Cirrhosis liver disease. The Enriched Seagull Optimization Algorithm (ESOA) is employed to fine-tune the parameters of the DNN approach. Lastly, the tumor is separated from the image using the suggested enhanced adaptive thresholding with a level-set segmentation (EATLSS). The experimental outcomes demonstrate that the proposed method achieved over 99.3% accuracy in both classification and segmentation tasks. The classification model outperforms existing approaches, achieving an accuracy of 99.40%, sensitivity of 99.32%, specificity of 98.56%, and precision of 98.99%. Likewise, the segmentation model delivers outstanding performance, with an accuracy of 99.47%, precision of 98.95%, and a Dice coefficient of 98.86%, highlighting its effectiveness in accurately delineating liver abnormalities.</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

OFT-DNN-EATLSS: An Enhanced Deep Learning-Based Approach for Efficient Cirrhosis Liver Disease Classification and Segmentation Using Multimodality Images

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • G. Sai Chaitanya Kumar,
  • Narendhar Mulugu,
  • M. Vijaya Kamal,
  • V. Srilakshmi,
  • G. N. Beena Bethel

摘要

The diffuse and advanced stage of liver disease is identified as liver cirrhosis. The detection of liver illness depends on the full automatic diagnosis procedure that identifies the liver and tumor regions from the image modalities. Moreover, many recent works have utilized a single image modality to classify and segment liver cirrhosis. These modalities, however, lack greater detection accuracy and are biased. Therefore, this work applies a novel model to improve the efficacy of liver tumor analysis by merging deep learning with image modalities. In this work, multimodality images (MMI), such as CT, MRI, and US, are taken into account. First, the noise in the input MMI is removed using a Median Filter (MF) and an Extended Guided Filter (EGF). Furthermore, contrast-limited adaptive histogram equalization (CLAHE), which maintains image brightness while enhancing original contrast, enhances the quality of liver multimodality images. After that, the Advanced Gray-Level Co-occurrence Matrix (AGLCM) facilitates the Feature Extraction (FE) function. To further improve accuracy, the key features were chosen using the Lotus Effect Optimization Algorithm (LEOA). After that, the proposed Optimal Fine-Tuned Deep Neural Network (OFT-DNN)-based deep learning model is used to categorize the Cirrhosis liver disease. The Enriched Seagull Optimization Algorithm (ESOA) is employed to fine-tune the parameters of the DNN approach. Lastly, the tumor is separated from the image using the suggested enhanced adaptive thresholding with a level-set segmentation (EATLSS). The experimental outcomes demonstrate that the proposed method achieved over 99.3% accuracy in both classification and segmentation tasks. The classification model outperforms existing approaches, achieving an accuracy of 99.40%, sensitivity of 99.32%, specificity of 98.56%, and precision of 98.99%. Likewise, the segmentation model delivers outstanding performance, with an accuracy of 99.47%, precision of 98.95%, and a Dice coefficient of 98.86%, highlighting its effectiveness in accurately delineating liver abnormalities.