Fully Automated Segmentation of Lower Extremity Deep Vein Thrombosis Using Convolutional Neural Network

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Deep vein thrombosis (DVT) is a serious condition requiring accurate diagnosis.
• Current DVT segmentation methods can be time-consuming and subjective.
• Automated segmentation can improve diagnostic efficiency and treatment planning.

Purpose of the Study:

• To develop and evaluate a fully automatic deep learning (DL) method for segmenting lower extremity deep vein thrombosis (DVT).
• To assess the performance of the proposed DL model using contrast-enhanced magnetic resonance imaging (CE-MRI).
• To compare the DL model's segmentation accuracy against other existing DL models.

Main Methods:

• A DL network with an encoder-decoder architecture was designed for DVT segmentation.
• CE-MRI images from 58 patients with lower extremity DVT were used.
• An 8-fold cross-validation strategy and Dice Similarity Coefficient (DSC) were employed for evaluation.

Main Results:

• The DL model achieved a mean DSC of 0.74 ± 0.17 and a median DSC of 0.79.
• Segmentation of a single MRI slice took approximately 1.5 seconds.
• The proposed CNN model outperformed other DL models with a DSC of 0.74 ± 0.17.

Conclusions:

• The developed DL method provides effective and rapid fully automatic segmentation of lower extremity DVT.
• This automated approach shows significant potential for clinical application in DVT diagnosis.
• The study highlights the efficacy of DL in medical image segmentation tasks.