In recent years, generative adversarial networks (GANs) have gained tremendous popularity for potential applications in medical imaging, such as medical image synthesis, restoration, reconstruction, translation, as well as objective image quality assessment. Despite the impressive progress in generating high-resolution, perceptually realistic images, it is not clear if modern GANs reliably learn the statistics that are meaningful to a downstream medical imaging application. In this work, the ability of a state-of-the-art GAN to learn the statistics of canonical stochastic image models (SIMs) that are relevant to objective assessment of image quality is investigated. It is shown that although the employed GAN successfully learned several basic first- and second-order statistics of the specific medical SIMs under consideration and generated images with high perceptual quality, it failed to correctly learn several per-image statistics pertinent to the these SIMs, highlighting the urgent need to assess medical image GANs in terms of objective measures of image quality.

We showed in our earlier work that the choice of reconstruction methods does not affect the optimization of DBT acquisition parameters (angular span and number of views) using simulated breast phantom images in detecting lesions with a channelized Hotelling observer (CHO). In this work we investigate whether the model-observer based conclusion is valid when using humans to interpret images. We used previously generated DBT breast phantom images and recruited human readers to find the optimal geometry settings associated with two reconstruction algorithms, filtered back projection (FBP) and simultaneous algebraic reconstruction technique (SART). The human reader results show that image quality trends as a function of the acquisition parameters are consistent between FBP and SART reconstructions. The consistent trends confirm that the optimization of DBT system geometry is insensitive to the choice of reconstruction algorithm. The results also show that humans perform better in SART reconstructed images than in FBP reconstructed images. In addition, we applied CHOs with three commonly used channel models, Laguerre–Gauss (LG) channels, square (SQR) channels and sparse difference-of-Gaussian (sDOG) channels. We found that LG channels predict human performance trends better than SQR and sDOG channel models for the task of detecting lesions in tomosynthesis backgrounds. Overall, this work confirms that the choice of reconstruction algorithm is not critical for optimizing DBT system acquisition parameters.

The availability of large medical image datasets is critical in many applications, such as training and testing of computer-aided diagnosis systems, evaluation of segmentation algorithms, and conducting perceptual studies. However, collection of data and establishment of ground truth for medical images are both costly and difficult. To address this problem, we are developing an image blending tool that allows users to modify or supplement existing datasets by seamlessly inserting a lesion extracted from a source image into a target image. In this study, we focus on the application of this tool to pulmonary nodules in chest CT exams. We minimize the impact of user skill on the perceived quality of the composite image by limiting user involvement to two simple steps: the user first draws a casual boundary around a nodule in the source, and, then, selects the center of desired insertion area in the target. We demonstrate the performance of our system on clinical samples, and report the results of a reader study evaluating the realism of inserted nodules compared to clinical nodules. We further evaluate our image blending techniques using phantoms simulated under different noise levels and reconstruction filters. Specifically, we compute the area under the ROC curve of the Hotelling observer (HO) and noise power spectrum of regions of interest enclosing native and inserted nodules, and compare the detectability, noise texture, and noise magnitude of inserted and native nodules. Our results indicate the viability of our approach for insertion of pulmonary nodules in clinical CT images.

Multi-slice computed tomography (MSCT) scanners have become popular volumetric imaging tools. Deterministic and random properties of the resulting CT scans have been studied in the literature. Due to the large number of voxels in the three-dimensional (3D) volumetric dataset, full characterization of the noise covariance in MSCT scans is difficult to tackle. However, as usage of such datasets for quantitative disease diagnosis grows, so does the importance of understanding the noise properties because of their effect on the accuracy of the clinical outcome. The goal of this work is to study noise covariance in the helical MSCT volumetric dataset. We explore possible approximations to the noise covariance matrix with reduced degrees of freedom, including voxel-based variance, one-dimensional (1D) correlation, two-dimensional (2D) in-plane correlation and the noise power spectrum (NPS). We further examine the effect of various noise covariance models on the accuracy of a prewhitening matched filter nodule size estimation strategy. Our simulation results suggest that the 1D longitudinal, 2D in-plane and NPS prewhitening approaches can improve the performance of nodule size estimation algorithms. When taking into account computational costs in determining noise characterizations, the NPS model may be the most efficient approximation to the MSCT noise covariance matrix.