Graph Attention Network based Pruning for Reconstructing 3D Liver Vessel Morphology from Contrasted CT Images (2003.07999v1)

Abstract: With the injection of contrast material into blood vessels, multi-phase contrasted CT images can enhance the visibility of vessel networks in the human body. Reconstructing the 3D geometric morphology of liver vessels from the contrasted CT images can enable multiple liver preoperative surgical planning applications. Automatic reconstruction of liver vessel morphology remains a challenging problem due to the morphological complexity of liver vessels and the inconsistent vessel intensities among different multi-phase contrasted CT images. On the other side, high integrity is required for the 3D reconstruction to avoid decision making biases. In this paper, we propose a framework for liver vessel morphology reconstruction using both a fully convolutional neural network and a graph attention network. A fully convolutional neural network is first trained to produce the liver vessel centerline heatmap. An over-reconstructed liver vessel graph model is then traced based on the heatmap using an image processing based algorithm. We use a graph attention network to prune the false-positive branches by predicting the presence probability of each segmented branch in the initial reconstruction using the aggregated CNN features. We evaluated the proposed framework on an in-house dataset consisting of 418 multi-phase abdomen CT images with contrast. The proposed graph network pruning improves the overall reconstruction F1 score by 6.4% over the baseline. It also outperformed the other state-of-the-art curvilinear structure reconstruction algorithms.

• The paper proposes a novel framework that integrates FCN-based heatmap generation with GAT for refining liver vessel graphs.
• It employs over-reconstruction followed by GAT-driven pruning to effectively eliminate false-positive vessel branches.
• The method achieved a 6.4% improvement in F1 score over baselines, enhancing the precision of preoperative surgical planning.

This paper addresses the complex challenge of automatically reconstructing the 3D morphological structure of liver vessels from contrasted CT images. This task is crucial for developing accurate liver preoperative surgical plans. The primary difficulties stem from the intricate shapes of liver vessels and varying intensities in different phases of CT images, which can affect visualization and analysis.

Methodology

The authors propose a novel framework that integrates both a Fully Convolutional Neural Network (FCN) and a Graph Attention Network (GAT) to enhance the reconstruction process:

1. Liver Vessel Centerline Detection:
   • An FCN is employed to generate a heatmap that indicates the centerlines of liver vessels. This serves as the preliminary step in identifying potential vessel regions.
2. Over-Reconstruction and Initial Graph Creation:
   • Utilizing image processing techniques, an initial liver vessel graph is constructed, tracing the paths suggested by the heatmap. This step tends to over-reconstruct, meaning it includes false-positive branches to ensure no potential vessel paths are missed.
3. Graph Attention Network for Pruning:
   • A GAT is then introduced to refine the initial graph. This network predicts the likelihood of each branch being a true vessel segment by aggregating features from the FCN. The GAT effectively prunes unnecessary paths, enhancing the accuracy of the model.

Evaluation and Results

The framework was tested on an in-house dataset comprising 418 multi-phase abdomen CT images with contrast. The proposed method significantly improved reconstruction accuracy:

• F1 Score Improvement: The inclusion of the graph network pruning enhanced the overall F1 score by 6.4% compared to a baseline method.
• Comparative Performance: The approach outperformed other state-of-the-art algorithms designed for curvilinear structure reconstruction, demonstrating its robustness and effectiveness.

Implications

The framework offers a substantial advancement in reconstructing liver vessel morphology, directly benefiting clinical applications by providing better tools for surgical planning. The integration of GAT for selective pruning also highlights the potential of graph-based methods in medical image analysis for improving the reliability and precision of anatomical reconstructions.

Overall, the paper presents a significant contribution to both the field of medical imaging and the application of graph neural networks in healthcare.