Automatic lymphedema segmentation in T2-STIR MRI using an unsupervised clustering method.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Quantitative Analysis

Background:

• Lymphedema and lipolymphedema assessment often relies on subjective interpretation of imaging.
• Accurate quantification of edema volume and distribution is crucial for effective management.

Purpose of the Study:

• To develop and validate an unsupervised AI method for automated segmentation of fluid in T2-STIR MRL.
• To quantitatively assess edema in patients with lymphedema and lipolymphedema.

Main Methods:

• Retrospective analysis of 20 patients with lymphedema or lipolymphedema.
• K-means clustering algorithm for image segmentation, optimized using Dice similarity coefficient.
• Transfer learning applied to a test set for performance evaluation and 3D analysis.

Main Results:

• The AI model achieved a Dice similarity coefficient of at least 0.8 on the training set.
• The model demonstrated good agreement with manual segmentation on the test set (Dice score 0.74 ± 0.05).
• Visualizations included color maps and stacked line plots for edema distribution and longitudinal tracking.

Conclusions:

• The unsupervised AI method shows potential for automated, objective edema segmentation and quantification in T2-STIR MRL.
• This approach can assist in the diagnosis, staging, and treatment monitoring of lymphedema and lipolymphedema.