Medical data processing apparatus, medical data processing method, and magnetic resonance imaging apparatus

Patent metadata

Field	Value
Publication number	US-2021003651-A1
Application number	US-202016912034-A
Country	US
Kind code	A1
Filing date	Jun 25, 2020
Priority date	Jul 3, 2019
Publication date	Jan 7, 2021
Grant date	—

Abstract

According to one embodiment, a medical data processing apparatus includes processing circuitry. The processing circuitry acquires first data pieces obtained by sparse sampling. The processing circuitry generates first compressed data pieces lower in number than the first data pieces by multiplying the first data pieces by each of sets of weight coefficients and adding each of the multiplied first data pieces. The processing circuitry performs first processing of outputting second compressed data pieces by applying a trained model to the first compressed data pieces, the trained model being trained by receiving first compressed data pieces based on sparse sampling and outputting at least one of second compressed data pieces based on full sampling.

First claim

What is claimed is:

1 . A medical data processing apparatus comprising processing circuitry configured to:
acquire a plurality of first data pieces obtained by sparse sampling; generate a plurality of first compressed data pieces lower in number than the plurality of first data pieces by multiplying the plurality of first data pieces by each of sets of weight coefficients and adding each of the multiplied first data pieces, the number of sets of weight coefficients being smaller than the number of the first data pieces; and perform first processing of outputting one or more second compressed data pieces by applying a trained model to the first compressed data pieces, the trained model being trained by receiving first compressed data pieces generated from first data pieces obtained by sparse sampling for an observation target and outputting at least one of plurality of second compressed data pieces generated from a plurality of second data pieces obtained by full sampling for the observation target.

2 . The apparatus of claim 1 , wherein the weight coefficients are determined based on coefficients obtained by compressing a dimension of the second data pieces.

3 . The apparatus of claim 1 , wherein the processing circuitry repeatedly performs the first processing and a second processing until a convergence condition is satisfied, the second processing correcting the first compressed data pieces based on a comparison between estimated first data pieces estimated from the second compressed data pieces and the first data pieces.

4 . The apparatus of claim 3 , wherein the processing circuitry repeatedly performs the first processing and the second processing, until the convergence condition is satisfied by using an alternating direction method of multipliers (ADMM).

5 . The apparatus of claim 1 , the processing circuitry estimates a quantitative value from the output second compressed data pieces.

6 . The apparatus of claim 1 , wherein the processing circuitry uses the weight coefficients generated by using dimensional compression method including singular value decomposition and principal component analysis.

7 . The apparatus of claim 1 , wherein the first data pieces and the second data pieces are one of image data, k-space data, sinogram space data or hybrid space data.

8 . The apparatus of claim 7 , wherein the first data pieces are decimation data in which slices are skipped in a slice selecting direction in k-space, and
the processing circuitry interpolates the decimation data by parallel imaging.

9 . The apparatus of claim 1 , wherein the pieces of first data are respectively a plurality of sets of data based on a signal acquired by respective coils;
the processing circuitry configured to:
generate the first compressed data for each of the sets of data; and
generate at least one of the pieces of second compressed data from the first compressed data of the sets in accordance with the trained model trained to generate the second compressed data from the pieces of first compressed data as an input.

10 . The medical data processing apparatus of claim 9 , wherein the sets of data are sets of data based on signals of virtual coils integrating a plurality of coils.

11 . A medical data processing method comprising:
acquiring a plurality of first data pieces obtained by sparse sampling; generating a plurality of first compressed data pieces lower in number than the plurality of first data pieces by multiplying the plurality of first data pieces by each of sets of weight coefficients and adding each of the multiplied first data pieces, the number of sets of weight coefficients being smaller than the number of the first data pieces; and performing first processing of outputting one or more second compressed data pieces by applying a trained model to the first compressed data pieces, the trained model being trained by receiving first compressed data pieces generated from first data pieces obtained by sparse sampling for an observation target and outputting at least one of plurality of second compressed data pieces generated from a plurality of second data pieces obtained by full sampling for the observation target.

12 . The method of claim 11 , wherein the weight coefficients are determined based on coefficients obtained by compressing a dimension of the second data pieces.

13 . The method of claim 11 , further comprising repeatedly performing the first processing and a second processing until a convergence condition is satisfied, the second processing correcting the first compressed data pieces based on a comparison between estimated first data pieces estimated from the second compressed data pieces and the first data pieces.

14 . The method of claim 13 , wherein the repeatedly performing the first processing and the second processing is performed until the convergence condition is satisfied by using an alternating direction method of multipliers (ADMM).

15 . The method of claim 11 , further comprising estimating a quantitative value from the output second compressed data pieces.

16 . The method of claim 11 , wherein the weight coefficients is generated by using dimensional compression method including singular value decomposition and principal component analysis.

17 . The method of claim 11 , wherein the first data pieces and the second data pieces are one of image data, k-space data, sinogram space data or hybrid space data.

18 . The method of claim 17 , wherein the first data pieces are decimation data in which slices are skipped in a slice selecting direction in k-space, and the method further comprising interpolating the decimation data by parallel imaging.

19 . The method of claim 1 , wherein the pieces of first data are respectively a plurality of sets of data based on a signal acquired by respective coils;
the method further comprising:
generating the first compressed data for each of the sets of data; and
generating at least one of the pieces of second compressed data from the first compressed data of the sets in accordance with the trained model trained to generate the second compressed data from the pieces of first compressed data as an input.

20 . A magnetic resonance imaging apparatus comprising:
a collection unit configured to collect a plurality of first magnetic resonance (MR) data pieces by imaging a subject in accordance with an imaging sequence corresponding to sparse sampling; and processing circuitry configured to:
generate a plurality of first compressed MR data pieces lower in number than the plurality of first MR data pieces by multiplying the plurality of first MR data pieces by each of sets of weight coefficients and adding each of the multiplied first MR data pieces, the number of sets of weight coefficients being smaller than the number of the first MR data pieces; and
perform first processing of outputting one or more second MR compressed data pieces by applying a trained model to the first compressed MR data pieces, the trained model being trained by receiving first compressed MR data pieces generated from first MR data pieces obtained by sparse sampling for an observation target and outputting at least one of plurality of second compressed MR data pieces generated from a plurality of second MR data pieces obtained by full sampling for the observation target.

Assignees

• Canon Medical Systems Corp

Inventors

• Kamiguchi Yuuzo

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/08

  Learning methods · CPC title