Rensselaer Polytechnic Institute

Hongming Shan

Ge Wang

Wenxiang Cong

Lars Arne Gjesteby

Qingsong Yang

Chenyu You

Guang Li

Juergen Hahn

Uwe Kruger

Huidong Xie

Mannudeep Kalra

A deep neural network for metal artifact reduction is described. A method for computed tomography (CT) metal artifact reduction (MAR) includes receiving a first CT image data; receiving a second CT image data; and generating, by an artificial neural network (ANN), CT output image data configured to include fewer artifacts compared to the first and second CT image data. The ANN includes at least two parallel CT image data streams and a CT output image data stream. A first of the at least two parallel CT image data streams is based, at least in part, on the first CT image data, a second of the at least two parallel CT image data stream is based, at least in part, on the second CT image data. The CT output image data stream is based, at least in part, on respective outputs of the at least two parallel CT image data streams.

Deep neural network for CT metal artifact reduction

A system for generating a high resolution (HR) computed tomography (CT) image from a low resolution (LR) CT image is described. The system includes a first generative adversarial network (GAN) and a second GAN. The first GAN includes a first generative neural network (G) configured to receive a training LR image dataset and to generate a corresponding estimated HR image dataset, and a first discriminative neural network (D) configured to compare a training HR image dataset and the estimated HR image dataset. The second GAN includes a second generative neural network (F) configured to receive the training HR image dataset and to generate a corresponding estimated LR image dataset, and a second discriminative neural network (D) configured to compare the training LR image dataset and the estimated LR image dataset. The system further includes an optimization module configured to determine an optimization function based, at least in part, on at least one of the estimated HR image dataset and/or the estimated LR image dataset. The optimization function contains at least one loss function. The optimization module is further configured to adjust a plurality of neural network parameters associated with at least one of the first GAN and/or the second GAN, to optimize the optimization function.

CT super-resolution GAN constrained by the identical, residual and cycle learning ensemble (GAN-circle)

A system for few-view computed tomography (CT) image reconstruction is described. The system includes a preprocessing module, a first generator network, and a discriminator network. The preprocessing module is configured to apply a ramp filter to an input sinogram to yield a filtered sinogram. The first generator network is configured to receive the filtered sinogram, to learn a filtered back-projection operation and to provide a first reconstructed image as output. The first reconstructed image corresponds to the input sinogram. The discriminator network is configured to determine whether a received image corresponds to the first reconstructed image or a corresponding ground truth image. The generator network and the discriminator network correspond to a Wasserstein generative adversarial network (WGAN). The WGAN is optimized using an objective function based, at least in part, on a Wasserstein distance and based, at least in part, on a gradient penalty.

Few-view CT image reconstruction system

Training a CNN with pseudo ground truth for CT artifact reduction is described. An estimated ground truth apparatus is configured to generate an estimated ground truth image based, at least in part, on an initial CT image that includes an artifact. Feature addition circuitry is configured to add a respective feature to each of a number, N, copies of the estimated ground truth image to create the number, N, initial training images. A computed tomography (CT) simulation circuitry is configured to generate a plurality of simulated training CT images based, at least in part, on at least some of the N initial training images. An artifact reduction circuitry is configured to generate a plurality of input training CT images based, at least in part, on the simulated training CT images. A CNN training circuitry is configured to train the CNN based, at least in part, on the input training CT images and based, at least in part, on the initial training images.

Training a CNN with pseudo ground truth for CT artifact reduction

A system for enhancing a low-dose (LD) computed tomography (CT) image is described. The system includes a modularized adaptive processing neural network (MAP-NN) apparatus and a MAP module. The MAP-NN apparatus is configured to receive a LDCT image as input. The MAP-NN apparatus includes a number, T, trained neural network (NN) modules coupled in series. Each trained NN module is configured to generate a respective test intermediate output image based, at least in part, on a respective received test input image. Each test intermediate output image corresponds to an incrementally denoised respective received test input image. The MAP module is configured to identify an optimum mapping depth, D, based, at least in part, on a selected test intermediate output image, the selected test intermediate output image selected by a domain expert. The mapping depth, D, is less than or equal to the number, T.

Modularized adaptive processing neural network (MAP-NN) for low-dose CT

A deep neural network for metal artifact reduction is described. A method for computed tomography (CT) metal artifact reduction (MAR) includes generating, by a projection completion circuitry, an intermediate CT image data based, at least in part, on input CT projection data. The intermediate CT image data is configured to include relatively fewer artifacts than an uncorrected CT image reconstructed from the input CT projection data. The method further includes generating, by an artificial neural network (ANN), CT output image data based, at least in part, on the intermediate CT image data. The CT output image data is configured to include relatively fewer artifacts compared to the intermediate CT image data. The method may further include generating, by detail image circuitry, detail CT image data based, at least in part, on input CT image data. The CT output image data is generated based, at least in part, on the detail CT image data.

A 3-D convolutional autoencoder for low-dose CT via transfer learning from a 2-D trained network is described, A machine learning method for low dose computed tomography (LDCT) image correction is provided. The method includes training, by a training circuitry, a neural network (NN) based, at least in part, on two-dimensional (2-D) training data. The 2-D training data includes a plurality of 2-D training image pairs. Each 2-D image pair includes one training input image and one corresponding target output image. The training includes adjusting at least one of a plurality of 2-D weights based, at least in part, on an objective function. The method further includes refining, by the training circuitry, the NN based, at least in part, on three-dimensional (3-D) training data. The 3-D training data includes a plurality of 3-D training image pairs. Each 3-D training image pair includes a plurality of adjacent 2-D training input images and at least one corresponding target output image. The refining includes adjusting at least one of a plurality of 3-D weights based, at least in part, on the plurality of 2-D weights and based, at least in part, on the objective function. The plurality of 2-D weights includes the at least one adjusted 2-D weight.

3-D convolutional autoencoder for low-dose CT via transfer learning from a 2-D trained network

Computed tomography (CT) screening, diagnosis, or another image analysis tasks are performed using one or more networks and/or algorithms to either integrate complementary tomographic image reconstructions and radiomics or map tomographic raw data directly to diagnostic findings in the machine learning framework. One or more reconstruction networks are trained to reconstruct tomographic images from a training set of CT projection data. One or more radiomics networks are trained to extract features from the tomographic images and associated training diagnostic data. The networks/algorithms are integrated into an end-to-end network and trained. A set of tomographic data, e.g., CT projection data, and other relevant information from an individual is input to the end-to-end network, and a potential diagnosis for the individual based on the features extracted by the end-to-end network is produced. The systems and methods can be applied to CT projection data, MRI data, nuclear imaging data, ultrasound signals, optical data, other types of tomographic data, or combinations thereof.

Growing community of inventors

Troy, NY, United States of America

Hongming Shan