Attention Connect Network for Liver Tumor Segmentation from CT and MRI Images

Area of Science:

• Medical Imaging and Artificial Intelligence
• Oncology and Radiology

Background:

• Liver cancer incidence is rising, necessitating accurate tumor identification for effective treatment.
• Manual liver tumor segmentation is time-consuming and subjective.
• Automated segmentation methods face challenges due to tumor variability.

Purpose of the Study:

• To develop an automated liver tumor segmentation method using an innovative deep learning approach.
• To improve the accuracy and efficiency of liver tumor contour identification in medical images.

Main Methods:

• Introduction of the attention connect network (AC-Net), a U-Net based architecture.
• Integration of axial attention module (AAM) and vision transformer module (VTM) to enhance feature integration.
• Pretraining on the LiTS2017 dataset followed by fine-tuning on CT and MRI data from Hubei Cancer Hospital.

Main Results:

• AC-Net achieved a Dice Similarity Coefficient (DSC) of 0.90 on CT data and 0.80 on MRI data.
• High precision (0.89 CT, 0.84 MRI) and recall (0.92 CT, 0.82 MRI) were reported for both modalities.
• Ablation studies confirmed the effectiveness of individual modules within AC-Net.

Conclusions:

• AC-Net demonstrates superior tumor recognition accuracy and competitive performance for clinical applications.
• The proposed attention modules significantly contribute to the network's segmentation capabilities.
• The study provides an effective automated solution for liver tumor segmentation.