Table 1.

Quantitative ultrasound biomarkers based on raw data.

Biomarker	Description	Literature references
Automatic morphometric measurements	Automatic muscle morphometric analysis based on neural network segmentation models trained with respect to sonographer annotations of rectus femoris cross section, and 2D Fourier analysis of pennation angle.	[72,73]
Attenuation coefficient	Measurement of loss of signal intensity with depth.	[47,48,50]
Backscattering coefficient	Measurement of tissue reflectivity after attenuation compensation.	[74,75]
Power spectrum (Lizzi-Feleppa parameters)	Spectroscopy measurement of backscattered signal variation with frequency, including parametrizations such as spectral slope, spectral intercept, and midband fit	[49]
Speckle statistics	Fitting of raw envelope signal to speckle statistical distribution models, including Rayleigh, Nakagami, and homodyned K-distribution. Estimation of scatterer concentration, spacing, and coherence from fitted model parameters.	[49,76,77]
Statistical moments	Nonparametric statistical moments capturing scatterer distribution and concentration, such as entropy, kurtosis, skewness, variance, anisotropy, and signal-to-noise ratio.	[78,79]
Coherence and speed of sound	Generalized spectrum analysis and estimation of coherence, mean scatterer spacing, and speed-of-sound in muscle.	[80,81]
Textural radiomics	First- and second-order texture features extracted from both B-mode (eg, based on gray-scale co-occurrence matrices) and raw data (eg, based on wavelet and Laplacian transformations) and combined with machine learning models trained with respect to clinical outcomes.	[51,53,54]
Artificial intelligence radiomics	Radiofrequency data and B-mode features extracted automatically with end-to-end neural network models trained with respect to clinical outcomes	[52,82-84]