Related Experiment Video
Updated: Aug 30, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Development of Novel Residual-Dense-Attention (RDA) U-Net Network Architecture for Hepatocellular Carcinoma
Wen-Fan Chen1, Hsin-You Ou2, Han-Yu Lin3
1Institute of Medical Science and Technology, National Sun Yat-sen University, Kaohsiung 80424, Taiwan.
Diagnostics (Basel, Switzerland)
|August 26, 2022
Summary
This study introduces a novel AI model for liver lesion segmentation using CT scans. The Residual-Dense-Attention U-Net achieves high accuracy and reduces computation time, aiding physician decision-making.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Deep Learning
Background:
- Accurate liver organ and lesion segmentation is crucial for diagnosis and treatment planning.
- Current methods may require contrast agents or face challenges with complex anatomical locations and lesion variations.
Purpose of the Study:
- To develop and evaluate an AI-based image recognition system for precise liver organ and lesion segmentation from CT images.
- To assess the performance of the proposed Residual-Dense-Attention (RDA) U-Net model, particularly its ability to function without contrast agents.
Main Methods:
- Utilized a hybrid U-Net architecture combining ResNet and DenseNet components in the encoder for efficient parameter retention.
- Incorporated Attention Gates in the decoder to focus on relevant image features and suppress irrelevant areas.
- Trained and validated the RDA U-Net model on a liver dataset from open-source (LiTS) and hospital sources using CT images.
Main Results:
- The RDA U-Net achieved high accuracy in segmenting liver organs (96%) and lesions (94.8%) without contrast agents.
- Demonstrated excellent segmentation across various liver locations and lesion types (large, small, single, multiple).
- Reduced overall computation time by approximately 28% compared to traditional convolutional methods, with strong IoU and AVGDIST scores.
Conclusions:
- The proposed RDA U-Net model demonstrates significant potential for accurate and efficient liver organ and lesion segmentation in medical imaging.
- This AI approach can assist physicians in clinical decision-making by providing reliable segmentation results, even without contrast enhancement.
- The model's efficiency and accuracy suggest its applicability in improving diagnostic workflows for liver pathologies.
