John Patrick Kaufhold

Thickness-dependent scatter correction algorithm for digital mammography

We have implemented a scatter-correction algorithm (SCA) for digital mammography based on an iterative restoration filter. The scatter contribution to the image is modeled by an additive component that is proportional to the filtered unattenuated x-ray photon signal and dependent on the characteristics of the imaged object. The SCAs result is closer to the scatter-free signal than when a scatter grid is used. Presently, the SCA shows improved contrast-to-noise performance relative to the scatter grid for a breast thickness up to 3.6 cm, with potential for better performance up to 6 cm. We investigated the efficacy of our scatter-correction method on a series of x-ray images of anthropomorphic breast phantoms with maximum thicknesses ranging from 3.0 cm to 6.0 cm. A comparison of the scatter-corrected images with the scatter-free signal acquired using a slit collimator shows average deviations of 3 percent or less, even in the edge region of the phantoms. These results indicate that the SCA is superior to a scatter grid for 2D quantitative mammography applications, and may enable 3D quantitative applications in X-ray tomosynthesis.

Registration and integration for fluoroscopy device enhancement

We investigated a method, motion compensated integration (MCI), for enhancing stent Contrast-to-Noise Ratio (CNR) such that stent deployment may be more easily assessed. MCI registers fluoroscopic frames on the basis of stent motion and performs pixel-wise integration to reduce noise. Registration is based on marker balls, high contrast interventional devices which guide the clinician in stent placement. It is assumed that stent motion is identical to that of the marker balls. Detecting marker balls and identifying their centroids with a high degree of accuracy is a non-trivial task. To address the required registration accuracy, in this work we examine MCIs visualization benefit as a function of registration error. We employ adaptive forced choice experiments to quantify human discrimination fidelity. Perception results are contrasted with CNR measurements. For each level of registration inaccuracy investigated, MCI conferred a benefit (p < 0.05) on stent deployment assessment suggesting the technique is tolerant of modest registration error. We also consider the blurring effect of cardiac motion during the x-ray pulse and select frames for integration as a function of cardiac phase imaged.

3D Visualization of X-ray Tomosynthesis Digital Mammography Data: Preference Study

A 3D visualization preference study was conducted on digital mammography tomosynthesis datasets. Two volume rendering [VR] techniques were used in the preference study, the Maximum Intensity Projection [MIP] and the Composite Ray Casting [CRC]. These techniques were presented side by side to experts in the field of digital mammography on dual high resolution monitors. The two techniques were presented using each of three different modes, the thick slice VR, tumbling along the azimuth direction, and tumbling along the elevation direction. Several phantoms and mastectomy specimens were imaged and reconstructed for this study. The experts ranked MIP higher than CRC, and tumbling along the elevation direction was selected as the best mode. In addition, the experts indicated that the MIP technique is better in displaying microcalcifications, and the CRC technique is better in showing masses.

Scatter Correction in Tomosynthesis Imaging for Mamography

Scatter is a significant source of bias and contrast-degradation in X-ray mammography. Accurately scatter-corrected projection images in tomosynthesis are expected to improve the performance of advanced 3D quantitative applications, such as tissue decomposition. In standard 2D mammography, scatter grids typically mitigate the scatter signal to an acceptable level. A scatter-rejection grid is not a practical option in X-ray tomosynthesis, where the X-ray source moves with respect to the detector. We have developed a scatter-correction algorithm based on inverse filtering, including a priori information about the tomosynthesis acquisition system. We investigated the performance of our scatter-correction method on a series of x-ray projection images of anthropomorphic breast phantoms, with thickness ranging from 3.0 cm to 6.0 cm, and acquired at angles ranging from 0 to 16 degrees with respect to the normal to the detector plane. A comparison of the scatter-corrected projection images with the scatter-free signal acquired using a slit collimator showed mean deviations of ˜3% or less for all investigated phantom thicknesses and source angles. Because our scatter-correction method is almost entirely automated, we have readily incorporated it into the 3D breast tomosynthesis reconstruction procedure. We present the effect of scatter correction on breast phantom tomosynthesis projections.

Longitudinal retrospective study of breast tissue composition

Links between subjective measurements of breast tissue composition and patient age have been demonstrated in the literature. Specifically, as women age, especially after menopause, radiologist-assigned breast tissue composition characterizations for a given patient regress from dense to less dense. Recently, a number of investigators have developed objective tools for quantitative breast tissue composition estimation. We evaluate the quantitative link between breast tissue composition and age, as well as other quantities, such as compressed breast height using the newer more quantitative methods for calculating breast tissue composition. We find that the compressed breast height is more strongly correlated with breast tissue composition than age. Further, we find that the quantitative tissue composition metric we calculate is correlated with the radiologist characterization of breast tissue composition.

Clinical Presentation of Breast Tissue Composition Metrics

Breast tissue composition is linked to both risk of developing breast cancer as well as the sensitivity of mammography. Thus, breast tissue composition assessment is important for patient management. However, it is not clear from published work if the visualization of breast tissue composition is relevant for clinical screening or diagnosis. We investigated the usefulness of different approaches to presenting quantitative tissue composition metrics to radiologists on a review workstation to address this question. Five different display methods were used to present the tissue composition metrics, 1) a word label, 2) a number, 3) superimposed contours, 4) a histogram, and 5) all together. For the specific presentation methods, we found that the radiologists polled preferred the first two methods to methods 3) and 4). All radiologists believed that it would be useful to present accurate, quantitative tissue composition information for establishing a consistent breast tissue composition score for each patient/mammogram.