Abstract
Dynamic contrast-enhanced (DCE) MRI has been used to quantitatively evaluate pulmonary perfusion based on the assumption of a gamma-variate function and an arterial input function (AIF) for deconvolution. However, these assumptions may be too simplistic and may not be valid in pathological conditions, especially in patients with complex inflow patterns (such as in congenital heart disease). Exploratory data analysis methods make minimal assumptions on the data and could overcome these pitfalls. In this work, two temporal clustering methods-Kohonen clustering network (KCN) and Fuzzy C-Means (FCM)-were concatenated to identify pixel time-course patterns. The results from seven normal volunteers show that this technique is superior for discriminating vessels and compartments in the pulmonary circulation. Patient studies with five cases of acquired or congenital pulmonary perfusion disorders demonstrate that pathologies can be highlighted in a concise map that combines information of the mean transit time (MTT) and pulmonary blood volume (PBV). The method was found to provide greater insight into the perfusion dynamics that might be overlooked by current model-based analyses, and may serve as a basis for optimal hemodynamic quantitative modeling of the interrogated perfusion compartments.
Original language | English |
---|---|
Pages (from-to) | 299-308 |
Number of pages | 10 |
Journal | Magnetic Resonance in Medicine |
Volume | 54 |
Issue number | 2 |
DOIs | |
State | Published - 08 2005 |
Externally published | Yes |
Keywords
- Data analysis
- Dynamic contrast-enhanced MRI
- Pathology
- Pulmonary perfusion
- Temporal clustering