Estimation of the biomechanical mammographic deformation of the breast using machine learning models.

BACKGROUND: A typical problem in the registration of MRI and X-ray mammography is the nonlinear deformation applied to the breast during mammography. We have developed a method for virtual deformation of the breast using a biomechanical model automatically constructed from MRI. The virtual deformation is applied in two steps: unloaded state estimation and compression simulation. The finite element method is used to solve the deformation process. However, the extensive computational cost prevents its usage in clinical routine. METHODS: We propose three machine learning models to overcome this problem: an extremely randomized tree (first model), extreme gradient boosting (second model), and deep learning-based bidirectional long short-term memory with an attention layer (third model) to predict the deformation of a biomechanical model. We evaluated our methods with 516 breasts with realistic compression ratios up to 76%. FINDINGS: We first applied one-fold validation, in which the second and third models performed better than the first model. We then applied ten-fold validation. For the unloaded state estimation, the median RMSE for the second and third models is 0.8 mm and 1.2 mm, respectively. For the compression, the median RMSE is 3.4 mm for both models. We evaluated correlations between model accuracy and characteristics of the clinical datasets such as compression ratio, breast volume, and tissue types. INTERPRETATION: Using the proposed models, we achieved accurate results comparable to the finite element model, with a speedup of factor 240 using the extreme gradient boosting model. These proposed models can replace the finite element model simulation, enabling clinically relevant real-time application.

Comparison of active-set method deconvolution and matched-filtering for derivation of an ultrasound transit time spectrum.

The quality of ultrasound computed tomography imaging is primarily determined by the accuracy of ultrasound transit time measurement. A major problem in analysis is the overlap of signals making it difficult to detect the correct transit time. The current standard is to apply a matched-filtering approach to the input and output signals. This study compares the matched-filtering technique with active set deconvolution to derive a transit time spectrum from a coded excitation chirp signal and the measured output signal. The ultrasound wave travels in a direct and a reflected path to the receiver, resulting in an overlap in the recorded output signal. The matched-filtering and deconvolution techniques were applied to determine the transit times associated with the two signal paths. Both techniques were able to detect the two different transit times; while matched-filtering has a better accuracy (0.13 µs versus 0.18 µs standard deviations), deconvolution has a 3.5 times improved side-lobe to main-lobe ratio. A higher side-lobe suppression is important to further improve image fidelity. These results suggest that a future combination of both techniques would provide improved signal detection and hence improved image fidelity.

Image fusion of Ultrasound Computer Tomography volumes with X-ray mammograms using a biomechanical model based 2D/3D registration.

Ultrasound Computer Tomography (USCT) is a promising breast imaging modality under development. Comparison to a standard method like mammography is essential for further development. Due to significant differences in image dimensionality and compression state of the breast, correlating USCT images and X-ray mammograms is challenging. In this paper we present a 2D/3D registration method to improve the spatial correspondence and allow direct comparison of the images. It is based on biomechanical modeling of the breast and simulation of the mammographic compression. We investigate the effect of including patient-specific material parameters estimated automatically from USCT images. The method was systematically evaluated using numerical phantoms and in-vivo data. The average registration accuracy using the automated registration was 11.9mm. Based on the registered images a method for analysis of the diagnostic value of the USCT images was developed and initially applied to analyze sound speed and attenuation images based on X-ray mammograms as ground truth. Combining sound speed and attenuation allows differentiating lesions from surrounding tissue. Overlaying this information on mammograms, combines quantitative and morphological information for multimodal diagnosis.

Automatic multimodal 2D/3D breast image registration using biomechanical FEM models and intensity-based optimization.

Due to their different physical origin, X-ray mammography and Magnetic Resonance Imaging (MRI) provide complementary diagnostic information. However, the correlation of their images is challenging due to differences in dimensionality, patient positioning and compression state of the breast. Our automated registration takes over part of the correlation task. The registration method is based on a biomechanical finite element model, which is used to simulate mammographic compression. The deformed MRI volume can be compared directly with the corresponding mammogram. The registration accuracy is determined by a number of patient-specific parameters. We optimize these parameters--e.g. breast rotation--using image similarity measures. The method was evaluated on 79 datasets from clinical routine. The mean target registration error was 13.2mm in a fully automated setting. On basis of our results, we conclude that a completely automated registration of volume images with 2D mammograms is feasible. The registration accuracy is within the clinically relevant range and thus beneficial for multimodal diagnosis.

Automatic image matching for breast cancer diagnostics by a 3D deformation model of the mamma.

X-ray mammograms and MR volumes provide complementary information for early breast cancer diagnosis. The breast is deformed during mammography, therefore the images can not be compared directly. A registration algorithm is investigated to fuse the images automatically. A finite element simulation was applied to a MR image of an underformed breast and compared to a compressed breast using different tissue models and boundary conditions. Based on the results a set of patient data was registered. To archive the requested accuracy distinguishing between the different tissue types of the breast was not necessary. A linear elastic model was sufficient. It was possible to simulate the deformation with an average deviation of approximately of the size of a voxel in the MRI data and retrieve the position of a lesion with an error of 3.8 mm in the patient data.

Elastic registration of x-ray mammograms and three-dimensional magnetic resonance imaging data.

Major problems in treating breast cancer are the early detection of tumors and accurate biopsy of small volumes of breast (mamma) tissue. This report presents an elastic registration algorithm of two x-ray mammograms and a corresponding magnetic resonance imaging (MRI) volume. To cope with the soft tissue deformation of the breast during mammography, a two-dimensional model of breast deformation behavior is used as an elastic transformation. Normalized mutual information is employed as a measure of similarity. Regions of interest in the uncompressed x-ray mammograms are projected into the MRI volume to determine their three-dimensional origin.