A double-gaussian, percentile-based method for estimating maximum blood flow velocity.

Area of Science:

• Neurosonology
• Biomedical Signal Processing
• Medical Imaging Analysis

Background:

• Transcranial Doppler (TCD) sonography estimates blood flow velocity, crucial for clinical assessment, especially peak systolic values.
• Current methods provide point estimates of maximum flow velocity, lacking information on signal uncertainty and distribution.
• This limitation is particularly problematic in conditions like cerebral vasospasm, where accurate measurement of high velocities is challenging.

Purpose of the Study:

• To develop a novel method for estimating blood flow velocity and its uncertainty using TCD data.
• To address the limitations of existing point-estimate methods in capturing the full picture of flow dynamics.
• To improve the clinical assessment of patients, particularly those with vasospasm.

Main Methods:

• A Gaussian mixture model was employed to differentiate signal from noise in TCD velocity data.
• Percentiles of the signal distribution were used to construct a flow velocity envelope.
• This approach naturally incorporates and visualizes uncertainty through multiple percentile curves (e.g., 95th, 99th).

Main Results:

• The developed Gaussian mixture model-based envelopes provided reasonable and useful estimates of maximal flow velocities in 59 patients.
• These new envelopes demonstrated comparable or superior utility compared to a standard algorithm.
• The commonly used envelope method was found to be generally consistent with the 90th percentile of the signal distribution.

Conclusions:

• Decomposing TCD flow velocity distributions into noise and signal components using a double-Gaussian mixture model is effective.
• Percentiles of the signal component offer meaningful quantification of maximal flow velocities and their associated uncertainty.
• This method enhances the reliability of TCD-based assessments, especially in complex hemodynamic states.