Disclosed herein is a medical instrument (100, 300, 400, 500) comprising: a memory (110) storing machine executable instructions (120) and a tomographic data assessment module (122) and a processor (106) configured for controlling the medical instrument. Execution of the machine executable instructions causes the processor to receive (200) measured tomographic data (124). The measured tomographic data is configured for being reconstructed into a tomographic image (308) of a subject (418). Execution of the machine executable instructions further causes the processor to receive (202) an image quality indicator (126, 126′, 126″) by inputting the measured tomographic data into the tomographic data assessment module. The tomographic data assessment module is configured for generating the image quality indicator in response to inputting the measured tomographic data. Execution of the machine executable instructions further causes the processor to provide (204) the image quality indicator to an operator using an operator signaling system (108).

In a computer-implemented method of training a reconstruction neural network algorithm used to reconstruct a Magnetic Resonance Imaging (MRI) image, a prediction of training MRI image is determined based on training MRI raw data and using the reconstruction neural network algorithm. A prediction of a presence or absence of the object in the training MRI image is determined based on the prediction of the training MRI image and using an object-detection algorithm. A loss value is determined based on a first difference between the ground truth of the training MRI image and the prediction of the training MRI image, and further based on a second difference between the ground truth of the presence or absence of the object and the prediction of the presence or absence of the object. Weights of the reconstruction neural network algorithm are adjusted based on the loss value and using a training process.

A method and apparatus is provided that uses a deep learning (DL) network to reduce noise and artifacts in reconstructed medical images, such as images generated using computed tomography, positron emission tomography, and magnetic resonance imaging. The DL network can operate either on pre-reconstruction data or on a reconstructed image. The DL network can be an artificial neural network or a convolutional neural network (e.g., using a three-channel volumetric kernel architecture). Different neural networks can be trained depending on the noise level, scanning protocol, or the anatomic, diagnostic or clinical objective of the reconstructed image (e.g., by partitioning the training data into noise-level range and training respective DL networks for each range). Further, the DL networks can be trained to mitigate artifacts, such as the cone-beam artifact.

A method for correcting projection images in CT image reconstruction is provided. The method may include obtaining a plurality of projection images of a subject. Each of the plurality of projection images may correspond to one of the plurality of gantry angles. The method may further include correcting a first projection image of the plurality of projection images according to a process for generating a corrected projection image. The process may include performing, based on the first projection image and a second projection image of the plurality of projection images, a first correction on the first projection image to generate a preliminary corrected first projection image. The process may also include performing, based on at least part of the preliminary corrected first projection image, a second correction on the preliminary corrected first projection image to generate a corrected first projection image corresponding to the first gantry angle.

A method is for generating image data of an examination object via a computed tomography system including a data processing unit; an X-ray radiation source and an X-ray radiation detector suspended on a support and mounted to be rotatable about a z-axis; and an examination table for supporting the examination object and a reference object arranged in a fixed position relative to the examination table. The method includes generating a raw data set by displacing the X-ray radiation source and the X-ray radiation detector relative to the examination object. During generation of the raw data set, at least one part of the examination object is sampled together with at least one part of the reference object. The sampling of the at least one part of the reference object is used to compensate at least in part for the influence of movement errors during the displacement.

The disclosure provides for a method for generating an ultrasound image that includes transmitting, by a plurality of transmitters in a transducer, at least two transmit beams at different angles, where at least parts of the transmit beams cover an overlapping region, and receiving, by a plurality of sensors of the transducer, reflected signals of the transmit beams. The method further comprises calculating channel coherence for the received signals to produce one or more channel coherence images, and calculating transmit coherence for the received signals to produce one or more transmit coherence images. The information from at least one of the channel coherence images and at least one of the transmit coherence images are combined to identify moving objects. The received signals from different transmits in overlapping regions are then processed to produce a final image that is compensated for the moving objects.

Disclosed is a three-dimensional low-dose PET reconstruction method based on deep learning. The method comprises the following steps: back projecting low-dose PET raw data to the image domain to maintain enough information from the raw data; selecting an appropriate three-dimensional deep neural network structure to fit the mapping between the back projection of the low-dose PET and a standard-dose PET image; after learning from the training samples the network parameters are fixed, realizing three-dimensional PET image reconstruction starting from low-dose PET raw data, thereby obtaining a low-dose PET reconstructed image which has a lower noise and a higher resolution compared with the traditional reconstruction algorithm and image domain noise reduction processing.

Proposed are a method and device for regularizing rapid three-dimensional tomographic imaging using a machine-learning algorithm. A method for regularizing three-dimensional tomographic imaging using a machine-learning algorithm according to an embodiment comprises the steps of: measuring a three-dimensional tomogram of a cell to acquire a raw tomogram of the cell; acquiring a regularized tomogram by using a regularization algorithm; and learning the relationship between the raw tomogram and the regularized tomogram through machine-learning.

In the scatter estimation method of the present invention, Step S1 (first TOF projection data generation) and Step S4 (non-TOF scatter estimation algorithm) are performed, and Step S2 (second TOF projection data generation) and Step S3 (calculation of TOF direction distribution ratio) are performed, and Step S5 (calculation of TOF scatter projection data) is performed. A distribution ratio is obtained from the second TOF projection data measured in a scattered radiation energy window (low energy window). Since the target of distribution is non-TOF scatter projection data in a reconstruction data energy window (standard energy window), post-distribution TOF scatter projection data is obtained as approximate TOF scatter projection data in the reconstruction data energy window (standard energy window), and scatter estimation can be accurately performed.

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.