Comparison of four attenuation-compensation methods for backscatter coefficient estimation and characterization of focal liver lesions

OBJECTIVE: To compare four attenuation-compensation methods for backscatter coefficient (BSC) estimation, assessment of contrast, and classification of focal liver lesions (FLL).&#xD;&#xD;Approach: Ninety-seven patients with 100 FLL were scanned to collect radiofrequency ultrasound images. BSC methods relied on a reference phantom for system and operator-settings independent estimations. Method #1 employed a priori tissue layer segmentation and documented attenuation coefficients (AC) of each layer. Method #2 used a fixed total AC (0.85 dB/cm/MHz). Method #3 used local AC compensation. Method #4 jointly estimated total AC and BSC with a power law frequency model (bf^η). BSC@3MHz, b, η, total AC slope, and total AC were computed within segmented lesions. Lesion contrast was assessed with the contrast-to-noise ratio (CNR) and classification performances were evaluated with the area under the receiver operating characteristic curve (AUC). The composite reference standard was a combination of MRI and histopathology.&#xD;&#xD;Main results: The study included 30 primary and 26 secondary cancers, and 44 benign nodules. Parameter b provided the highest CNRs among BSC parameters (p < 0.0001) and gave higher CNRs than B-mode images (p < 0.0001). Method #3 was unsuitable with out-of-range values. Methods #1, #2, and #4 showed no significant differences for b, η, total AC slope, and total AC, whereas BSC@3MHz showed overestimations with Method #4 compared with Methods #1 and #2 (p < 0.001 and p < 0.0001, respectively). For differentiating benign and malignant lesions, η provided the highest AUC of 0.73 (95% confidence interval (CI): 0.62-0.82). For differentiating primary and secondary cancers, BSC@3MHz provided the highest AUC of 0.71 (95% CI: 0.55-0.83). Classification AUCs did not differ between Methods #1, #2 and #4.&#xD;&#xD;Significance: BSC imaging improved lesion contrast compared to B-mode and could classify FLL. Method #4 emerged as the most practical as it does not require any a priori AC or segmentation.&#xD.