A human observer study for evaluation and optimization of reconstruction methods in breast tomosynthesis using clinical cases

In breast tomosynthesis1 (BT) a number of 2D projection images are acquired from different angles along a limited arc. The imaged breast volume is reconstructed from the projection images, providing 3D information. The purpose of the study was to investigate and optimize different reconstruction methods for BT in terms of image quality using human observers viewing clinical cases. Sixty-six cases with suspected masses and calcifications were collected from 55 patients. Four different reconstructions of each image set were evaluated by four observers (two experienced radiologists, two experienced medical physicists): filtered back projection (FBP), iterative adapted FBP (iFBP) and two ML-convex iterative algorithm (MLCI) reconstructions (8 and 10 iterations) that differed in noise level and contrast of clinical details. Representation of masses and microcalcifications was evaluated. The structures were rated according to the overall appearance in a rank-order study. The differently reconstructed images of the same structure were displayed side by side in random order. The observers were forced to rank the order of the different reconstructed images and their proportions at each rank were scored. The results suggest that even though the FBP contains most noise its reconstructions are considered best overall, followed by iFBP, which contains least noise. In both FBP and iFBP methods the sharp borders and mass speculations were better represented than in iterative reconstructions while out-of-plane artifacts were better suppressed in the latter. However, in clinical practice the differences between the reconstructions may be considered negligible