Time-domain and 3D methods for lung perfusion data clustering in electrical impedance tomography

Area of Science:

• Medical Imaging
• Physiological Monitoring
• Computational Biology

Background:

• Pulmonary perfusion assessment is crucial for diagnosing lung conditions.
• Electrical Impedance Tomography (EIT) offers a non-invasive method for monitoring lung function.
• Accurate image segmentation is vital for reliable EIT-based perfusion analysis.

Purpose of the Study:

• To evaluate clustering methods for segmenting EIT images in pulmonary perfusion studies.
• To differentiate between lung pixels and hybrid pixels affected by heart and lung volume effects.
• To identify the most effective clustering algorithm for accurate perfusion quantification.

Main Methods:

• Utilized data from 51 perfusions in 8 swine (healthy and injured, mechanically ventilated).
• Developed ground truth masks for evaluating segmentation accuracy.
• Compared various clustering methods, including k-means with different metrics.

Main Results:

• The k-means method with the correlation metric showed the most effective and balanced performance.
• Achieved a median sensitivity and specificity of 84%.
• Effectively minimized false negatives, preventing misattribution of perfusion values.

Conclusions:

• K-means clustering with a correlation metric is a promising tool for EIT image segmentation in pulmonary perfusion studies.
• Accurate segmentation is essential for reliable interpretation of lung perfusion data.
• This method aids in distinguishing true lung perfusion from artifacts in EIT imaging.