Canon Medical Systems Corporation

Tzu-Cheng Lee

Zhou Yu

Jian Zhou

Liang Cai

Hiroki Taguchi

Yujie Lu

Masakazu Matsuura

Takuya Nemoto

Ilmar A Hein

Richard Thompson

Xi Chen

A method of denoising a plurality of input images, including grouping the plurality of input images into a plurality of groups; determining, for each group of the plurality of groups, an initial representative image for the group; performing, for each group, a sequence of T1 denoising sampling steps using a diffusion-based probabilistic model, starting with the initial representative image for the group, to generate a corresponding sequence of representative images for the group, wherein T1 is an integer greater than or equal to 1; and for each input image in each group, performing a second sequence of T2 denoising sampling steps using the model to generate a final image that is a restored image corresponding to the input image, wherein T2 is an integer greater than or equal to 1, and T=T1+T2.

Fast diffusion-based image restoration workflow via sharing of initial diffusion steps

A medical data processing method according to an embodiment includes inputting first medical data relating to a subject imaged with a medical image capture apparatus to a learned model to configured to generate second medical data having lower noise than that of the first medical data and having a super resolution compared with the first medical data based on the first medical data to output the second medical data.

Medical data processing method, model generation method, medical data processing apparatus, and computer-readable non-transitory storage medium storing medical data processing program

A medical image processing method includes obtaining a first set of projection data by performing, with a first CT apparatus including a first detector with a first pixel size, a first CT scan of an object in a first imaging region of the first detector; obtaining a first CT image with a first resolution by reconstructing the first set of projection data; obtaining a processed CT image with a resolution higher than the first resolution by applying a machine-learning model for resolution enhancement to the first CT image; and displaying the processed CT image or outputting the processed CT image for analysis. The machine-learning model is obtained by training using a second CT image based on a second set of projection data acquired by a second CT scan of the object in a second imaging region with a second CT apparatus including a second detector with a second pixel size.

Medical image processing method, medical image processing apparatus, and computer readable non-volatile storage medium storing medical image processing program

Devices, systems, and methods obtain first radiographic-image data reconstructed based on a set of projection data acquired in a radiographic scan; apply one or more trained machine-learning models to the set of projection data and the first radiographic-image data to obtain a set of parameters for a scatter kernel; input the set of parameters and the set of projection data into the scatter kernel to obtain scatter-distribution data; and perform scatter correction on the set of projection data using the scatter-distribution data, to obtain a set of corrected projection data.

Devices, systems, and methods for deep-learning kernel-based scatter estimation and correction

A method and apparatus is provided that uses a deep learning (DL) network together with a multi-resolution detector to perform X-ray projection imaging to provide improved resolution similar to a single-resolution detector but at lower cost and less demand on the communication bandwidth between the rotating and stationary parts of an X-ray gantry. The DL network is trained using a training dataset that includes input data and target data. The input data includes projection data acquired using a multi-resolution detector, and the target data includes projection data acquired using a single-resolution, high-resolution detector. Thus, the DL network is trained to improve the resolution of projection data acquired using a multi-resolution detector. Further, the DL network is can be trained to additional correct other aspects of the projection data (e.g., noise and artifacts).

Apparatus and method combining deep learning (DL) with an X-ray computed tomography (CT) scanner having a multi-resolution detector

A method and apparatus is provided to improve the image quality of images generated by analytical reconstruction of a computed tomography (CT) image. This improved image quality results from a deep learning (DL) network that is used to filter a sinogram before back projection but after the sinogram has been filtered using a ramp filter or other reconstruction kernel.

Apparatus and method using deep learning (DL) to improve analytical tomographic image reconstruction

A method and apparatus is provided that uses a deep learning (DL) network to correct projection images acquired using an X-ray source with a large focal spot size. The DL network is trained using a training dataset that includes input data and target data. The input data includes large-focal-spot-size X-ray projection data, and the output data includes small-focal-spot-size X-ray projection data (i.e., smaller than the focal spot of the input data). Thus, the DL network is trained to improve the resolution of projection data acquired using a large focal spot size, and obtain a resolution similar to what is achieved using a small focal spot size. Further, the DL network is can be trained to additional correct other aspects of the projection data (e.g., denoising the projection data).

Apparatus and method using deep learning (DL) to compensate for large focal spot size in x-ray projection imaging

A method and apparatus is provided to reduce the noise in medical imaging by training a deep learning (DL) network to select the optimal parameters for a convolution kernel of an adaptive filter that is applied in the data domain. For example, in X-ray computed tomography (CT) the adaptive filter applies smoothing to a sinogram, and the optimal amount of the smoothing and orientation of the kernel (e.g., a bivariate Gaussian) can be determined on a pixel-by-pixel basis by applying a noisy sinogram to the DL network, which outputs the parameters of the filter (e.g., the orientation and variances of the Gaussian kernel). The DL network is trained using a training data set including target data (e.g., the gold standard) and input data. The input data can be sinograms generated by a low-dose CT scan, and the target data generated by a high-dose CT scan.

Apparatus and method for sinogram restoration in computed tomography (CT) using adaptive filtering with deep learning (DL)

A method and apparatus are provided for using a neural network to estimate scatter in X-ray projection images and then correct for the X-ray scatter. For example, the neural network is a three-dimensional convolutional neural network 3D-CNN to which are applied projection images, at a given view, for respective energy bins and/or material components. The projection images can be obtained by material decomposing spectral projection data, or by segmenting a reconstructed CT image into material-component images, which are then forward projected to generate energy-resolved material-component projections. The result generated by the 3D-CNN is an estimated scatter flux. To train the 3D-CNN, the target scatter flux in the training data can be simulated using a radiative transfer equation method.

Growing community of inventors

Vernon Hills, IL, United States of America

Tzu-Cheng Lee