Hepatic vessels segmentation using deep learning and preprocessing enhancement

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate segmentation of liver hepatic vessels is vital for diagnosing hepatic diseases and planning surgical treatments.
• Convolutional Neural Networks (CNNs) have shown significant promise in medical image segmentation tasks.

Purpose of the Study:

• To develop an automatic deep learning system for segmenting liver hepatic vessels in Computed Tomography (CT) datasets.
• To improve the accuracy of liver vessel segmentation for enhanced preoperative planning.

Main Methods:

• A U-net based deep learning architecture with modified residual blocks and concatenation skip connections was implemented.
• Preprocessing steps included Coherence Enhancing Diffusion (CED) filtering and vesselness filtering to enhance vessel visibility.
• The impact of filtering enhancement and data mismatch during training and validation was investigated.

Main Results:

• The proposed system achieved an average Dice Similarity Coefficient (DSC) score of 79% on various CT datasets.
• The study evaluated the effectiveness of preprocessing filters and data handling strategies.

Conclusions:

• The developed deep learning approach accurately segments liver vasculature within the liver envelope.
• This automated segmentation method shows potential as a valuable tool for clinical preoperative planning in liver surgery.