Association of automated and human observer lesion detecting ability using phantoms

A set of tissue-mimicking phantoms containing spherical negative contrast simulated lesions was employed to associate an automated method for determining detectability with human observers. Six alternative methods for computing the lesion signal-to-noise ratio (LSNR) were employed for quantifying automated detecting ability. The six methods differ regarding effective lesion area and whether or not gradients in local mean background echo levels were accounted for. The two-alternative-forced-choice (TAFC) technique was used to associate detecting ability of human observers with LSNR values. Although the six methods gave similar results, one method exhibited the least dependency on lesion diameter and is recommended; that method accounts for gradients in local mean background echo levels and employs an effective sphere area of 2/π times the projected sphere area. A reasonable LSNR detection threshold value of −2.0 was found to apply for nominal transducer frequencies from 4 through 6 MHz and for lesion diameters from 2 through 5 mm. This result allows rapid human-observer-calibrated automated determination of the depth range of detectability as a function of sphere diameter and contrast. (E-mail: elmadsen@wisc.edu)