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

Patent metadata

Field	Value
Publication number	US-12299841-B2
Application number	US-202217705030-A
Country	US
Kind code	B2
Filing date	Mar 25, 2022
Priority date	Apr 7, 2021
Publication date	May 13, 2025
Grant date	May 13, 2025

Abstract

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.

First claim

What is claimed is:

1. A medical data processing method, comprising:
inputting first medical data relating to a subject imaged with a medical image capture apparatus to a learned model;
outputting second medical data having lower noise than the first medical data and having a higher resolution than the first medical data;
performing reconstruction to generate a first training image based on first pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the second reconstructed image;
adding noise to the first pre-reconstruction data and lowering a resolution of the first pre-reconstruction data to generate second pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the first reconstructed image;
performing reconstruction to generate a second training image based on the second pre-reconstruction data; and
training a deep convolution neural network using the first training image and the second training image to generate the learned model, wherein
the learned model has been trained using first training data and second training data, the first training data relating to a subject imaged with a medical image capture apparatus, and the second training data being generated by adding noise to the first training data for training the learned model and lowering a resolution of the first training data, and
the learned model has been learned to generate the first training data based on the second training data, the generated first training data having lower noise than the second training data and having a higher resolution than the second training data.

2. The medical data processing method according to claim 1 , wherein the first medical data is data before reconstruction or data before display processing, collected by imaging the subject with the medical image capture apparatus, and
the method further includes generating a medical image based on the second medical data.

3. The medical data processing method according to claim 1 , wherein
the first medical data is a first reconstructed image reconstructed based on collected data collected by imaging the subject with the medical image capture apparatus, and
the second medical data is a second reconstructed image having lower noise than that of the first reconstructed image and having a super resolution compared with the first reconstructed image.

4. The medical data processing method according to claim 3 , further comprising:
when the learned model is not used, performing reconstruction to generate the first reconstructed image in a first matrix size based on the collected data collected by imaging the subject with the medical image capture apparatus;
when the learned model is used, performing reconstruction to generate the first reconstructed image in a second matrix size based on the collected data, the second matrix size being greater than the first matrix size and coinciding with a matrix size of the second reconstructed image; and
inputting the first reconstructed image having the second matrix size to the learned model to output the second reconstructed image.

5. The medical data processing method according to claim 3 , further comprising:
upsampling the first reconstructed image to a second size greater than a first size and equal to the second reconstructed image; and
inputting the first reconstructed image having the second size to the learned model to output the second reconstructed image.

6. The medical data processing method of generating the learned model according to claim 2 , the medical data processing method comprising:
the first training data corresponding to the noise and the resolution of the second medical data, the second training data corresponding to noise and a resolution of the collected data; and
training a deep convolution neural network using the first training data and the second training data to generate the learned model.

7. A medical data processing method comprising:
inputting first medical data relating to a subject imaged with a medical image capture apparatus to a learned model;
outputting second medical data having lower noise than the first medical data and having a higher resolution than the first medical data;
performing reconstruction to generate a first training image based on first pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the second reconstructed image;
adding noise to the first pre-reconstruction data and reconstructing the first pre-reconstruction data to generate a noise-added image that corresponds to the noise of the first reconstructed image;
lowering a resolution of the noise-added image to generate a second training image that corresponds to the noise and the resolution of the first reconstructed image; and
training a deep convolution neural network using the first training image and the second training image to generate the learned model.

8. The medical data processing method of generating the learned model according to claim 3 , the medical data processing method comprising:
performing reconstruction to generate a first training image based on first pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the second reconstructed image;
lowering the resolution of the first pre-reconstruction data and reconstructing the first pre-reconstruction data to generate a low-resolution image that corresponds to the resolution of the first reconstructed image;
adding noise to the low-resolution image to generate a second training image that corresponds to the noise and the resolution of the first reconstructed image; and
training a deep convolution neural network using the first training image and the second training image to generate the learned model.

9. The medical data processing method of generating the learned model according to claim 3 , the medical data processing method comprising:
performing reconstruction to generate a first training image based on first pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the second reconstructed image;
adding noise to the first training image and lowering a resolution of the first training image to generate a second training image that corresponds to the noise and the resolution of the first reconstructed image; and
training a deep convolution neural network using the first training image and the second training image to generate the learned model.

10. A medical data processing apparatus comprising:
a processor configured to input first medical data relating to a subject imaged with a medical image capture apparatus to a learned model; and
the processor configured to output second medical data having lower noise than the first medical data and having a higher resolution than the first medical data;
the processor configured to perform reconstruction to generate a first training image based on first pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the second reconstructed image;
the processor configured to add noise to the first pre-reconstruction data and lowering a resolution of the first pre-reconstruction data to generate second pre-reconstruction data before reconstruction that corresponds to the noise and the resolution of the first reconstructed image;
the processor configured to perform reconstruction to generate a second training image based on the second pre-reconstruction data; and
the processor configured to train a deep convolution neural network using the first training image and the second training image to generate the learned model, wherein
the learned mo

Assignees

• Canon Medical Systems Corp

Inventors

• Matsuura Masakazu
• Nemoto Takuya
• Taguchi Hiroki
• Lee Tzu-Cheng
• Zhou Jian
• Cai Liang
• Yu Zhou

Classifications

• G06T12/30

  Image post-processing, e.g. metal artefact correction · CPC title

• G06T5/70

  Denoising; Smoothing · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20016

  Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform · CPC title