Metastatic liver tumour segmentation with a neural network-guided 3D deformable model

Area of Science:

• Medical Imaging
• Machine Learning
• Computational Biology

Background:

• Accurate liver tumor segmentation in CT images is crucial for cancer diagnosis, treatment planning, and response evaluation.
• Manual segmentation is time-consuming, often leading to estimations rather than precise volume assessments.
• Existing methods lack efficiency and accuracy when dealing with variable image resolutions.

Purpose of the Study:

• To develop and evaluate a semi-automatic segmentation method for liver tumors in CT images.
• To enhance the accuracy and efficiency of tumor volume assessment.
• To provide a robust tool for clinical use in liver cancer management.

Main Methods:

• A semi-automatic segmentation approach integrating machine learning with a deformable surface model.
• Utilizing a multilayer perceptron (MLP) based voxel classifier to interpret CT image data.
• Incorporating vertex displacement towards tumor boundaries and surface regularization for smoothness.

Main Results:

• The method was tested on 40 abdominal CT scans with 95 colorectal metastases from diverse scanners.
• Achieved encouraging segmentation results with a Dice similarity metric of [Formula: see text].
• Demonstrated capability in handling highly variable CT image data and resolutions.

Conclusions:

• The proposed semi-automatic method offers a promising solution for accurate liver tumor segmentation.
• Machine learning integrated with deformable models can effectively address challenges in medical image analysis.
• Further research with larger datasets and advanced neural networks is warranted to optimize performance.