A physically based, probabilistic model for ultrasonic images incorporating shape, microstructure, and system...

Area of Science:

• Medical image analysis
• Biophysics
• Computational imaging

Background:

• Bayesian methods show promise in medical image shape analysis.
• Existing probabilistic ultrasound models rely on analytic forms (e.g., Rayleigh, Rician) and intensive computation, limiting their applicability.
• A validated physical model for ultrasound image formation, incorporating system characteristics, surface shape, and microstructure, has been previously described.

Purpose of the Study:

• To develop a computationally feasible, pixel-based, probabilistic image model for ultrasound.
• To create a shape-based data likelihood for improved medical image analysis.
• To overcome limitations of existing analytic probabilistic models in ultrasound imaging.

Main Methods:

• A random phasor sum representation of the physical model was used to derive a probabilistic form.
• Amplitude mean and variance were computed directly from the physical model, unlike previous methods.
• A data likelihood was constructed by assigning density functions (Rayleigh and Gaussian) to each pixel based on computed statistics (SNR0).

Main Results:

• The new model allows direct computation of image statistics (mean, variance) from the physical model.
• This approach enables the inclusion of local surface shape, microstructure, and system characteristics at each pixel.
• A data likelihood was successfully constructed using a product of Rayleigh and Gaussian density functions based on pixel-wise SNR0 assessment.

Conclusions:

• The developed probabilistic model offers a computationally feasible and physically grounded approach for ultrasound image analysis.
• This method enhances the accuracy of data likelihood construction by incorporating detailed image formation parameters.
• The findings support the use of this shape-based probabilistic model for advanced medical image analysis in ultrasound.