Quantitative Evaluation of Image Quality for Dedicated Cone-Beam Breast Computerized Tomography

Abstract

Breast cancer is one of the common cancers in woman and early detection reduces the mortality rates. Offset-detector Cone Beam Breast Computerized Tomography (CBBCT) has been developed to improve breast cancer detection. Quantitative imaging metrices including modulation transfer function (MTF), noise power spectrum (NPS), contrast to noise ratio (CNR), and threshold detectability were studied to optimize the system and minimize radiation dose. A breast phantom with three modules (low contrast lesions, microcalcification specks, and homogeneous region) was scanned three times using tube current range of 16-100 mA with 49 kV. The projection datasets were reconstructed using a filtered-backprojection algorithm with modified Shepp-Logan filter. The disk objects in the low contrast module were used to determine MTF. The edge spread function was determined by fitting a sigmoid curve and differentiating it to obtain the line spread function. The limiting resolution (10% MTF) at the center of the field of view was found to be 2.4508 ± 0.0245 cycles/mm, corresponding to smallest detectable object size of 204 µm. The NPS was obtained from the homogenous module at different tube currents and at different slice thicknesses. Thirty-six ROIs were selected for each setting, and the background trend was subtracted with each ROI to obtain zero-mean image. 2DNPS was radially averaged to obtain 1DNPS and integration of 2DNPS yielded noise variance. As slice thickness or tube current increases, 1D NPS decreases while maintaining shape. Noise variance decreases exponentially with increasing mean glandular dose or slice thickness. The CNR and threshold detectability were studied using different tube currents, and hence mean glandular dose (MGD). CNR was determined from the low contrast module region, and the results showed linearity with MGD. Threshold detectability for both the low contrast module and the calcification specks were evaluated by human observer. With the radiation dose at screening level (4–6 mGy), the 270 μm calcification cluster and the 3 mm soft tissue lesion were detected. When radiation dose is increased to diagnostic imaging (~10 mGy), threshold detectability improved to 220–240 μm for calcification cluster and 2–3 mm for soft tissue lesion. The determined image quality in this work can be used to adjustment of acquisition parameters and the design of reconstruction algorithms.