Search

A fast and robust image registration algorithm for high-dimensional brain Magnetic Resonance images was developed based on the statistical deformation models (SDMs). This model learns deformation fields and achieves fast and robust registration by greatly reducing transformation dimensionality. This model is trained via principal component analysis (PCA), which suffers from large transformation dimensionality and small samples. For the high-dimensional image registration, the dimensions of the deformation fields are huge, the basic functions computed from PCA cannot represent deformation fields well. Therefore, we proposed a local SDM (LSDM) in this paper to solve the aforementioned problems. We divided the images into several small parts, in which the dimensions of the deformation fields are greatly reduced. Then, we trained the LSDM using the deformation fields between sample images and a selected template by applying PCA in each small part. Given that the dimension of eigenvectors of LSDM decreases much more than that of SDM, the orthonormal basis functions of LSDM represent the deformation fields more accurately than those of SDM. We obtained the total deformation fields for warping the image by integrating the deformation fields of all LSDMs. Using the manually labeled MR images of different people, we demonstrated that LSDM could greatly reduce the image registration time while maintaining favorable registration accuracy. [ABSTRACT FROM AUTHOR]

• Multi-input convolutional neural network for robust surface detection • Pose estimation integrated with spatial calibration for accurate tracking of ultrasound scans • First 3-D rendering of the full lumbar spine surface from free-hand untracked ultrasound scans using a rabbit animal model in vivo Three-dimensional (3-D) reconstruction of the spine surface is of strong clinical relevance for the diagnosis and prognosis of spine disorders and intra-operative image guidance. In this paper, we report a new technique to reconstruct lumbar spine surfaces in 3-D from non-invasive ultrasound (US) images acquired in free-hand mode. US images randomly sampled from in vivo scans of 9 rabbits were used to train a U-net convolutional neural network (CNN). More specifically, a late fusion (LF)-based U-net trained jointly on B-mode and shadow-enhanced B-mode images was generated by fusing two individual U-nets and expanding the set of trainable parameters to around twice the capacity of a basic U-net. This U-net was then applied to predict spine surface labels in in vivo images obtained from another rabbit, which were then used for 3-D spine surface reconstruction. The underlying pose of the transducer during the scan was estimated by registering stacks of US images to a geometrical model derived from corresponding CT data and used to align detected surface points. Final performance of the reconstruction method was assessed by computing the mean absolute error (MAE) between pairs of spine surface points detected from US and CT and by counting the total number of surface points detected from US. Comparison was made between the LF-based U-net and a previously developed phase symmetry (PS)-based method. Using the LF-based U-net, the averaged number of US surface points across the lumbar region increased by 21.61% and MAE reduced by 26.28% relative to the PS-based method. The overall MAE (in mm) was 0.24 ± 0.29. Based on these results, we conclude that: 1) the proposed U-net can detect the spine posterior arch with low MAE and large number of US surface points and 2) the newly proposed reconstruction framework may complement and, under certain circumstances, be used without the aid of an external tracking system in intra-operative spine applications. [Display omitted] [ABSTRACT FROM AUTHOR]

This paper proposes an infrared target detection algorithm based on lateral inhibition (LI) and singular value decomposition (SVD). Firstly, a local structure descriptor based on SVD of gradient domain is constructed, which reflects basic structures of the local regions of an infrared image. Then, LI network is modified by combining LI with the local structure descriptor for enhancing target and suppressing background. Meanwhile, to calculate lateral inhibition coefficients adaptively, the direction parameters are determined by the dominant orientations obtained from SVD. Experimental results show that, compared with the typical algorithms, the proposed algorithm not only can detect small target or area target under complex backgrounds, but also has excellent abilities of background suppression and target enhancement. [ABSTRACT FROM AUTHOR]

Medical ultrasound images are corrupted by speckle noise, which is multiplicative. This noise limits the contrast resolution in these images and complicates image-based quantitative measurement and diagnosis. In this study, the speckle noise in the ultrasound image is modeled by local statistics of the intensity distribution. And the non-local mean (NLM) filter is utilized to filter additional noise by applying the redundancy information in noisy images. A hybrid denoising method is proposed in consideration of the characteristics of both the local statistics of speckle noise and the NLM filter. The study combines local statistics with the NLM filter to reduce speckle in ultrasound images. The local statistics of speckle noise is estimated by local patches, while the intensity of the denoising pixel is computed by the weighted average of all the pixels by using the NLM. The weight is determined according to the similarity measures between the intensities of the local patches. The performance of the proposed method is evaluated on synthetic data, simulated images, and real images. Results of quantitative analysis and visual inspection of the synthetic data and of the real images demonstrate that the proposed method outperforms the original NLM, as well as many previously developed methods. [ABSTRACT FROM AUTHOR]