Medical image synthesis of abnormality patterns associated with COVID-19

Patent metadata

Field	Value
Publication number	US-11810291-B2
Application number	US-202016865266-A
Country	US
Kind code	B2
Filing date	May 1, 2020
Priority date	Apr 15, 2020
Publication date	Nov 7, 2023
Grant date	Nov 7, 2023

Abstract

Systems and methods for generating a synthesized medical image are provided. An input medical image is received. A synthesized segmentation mask is generated. The input medical image is masked based on the synthesized segmentation mask. The masked input medical image has an unmasked portion and a masked portion. An initial synthesized medical image is generated using a trained machine learning based generator network. The initial synthesized medical image includes a synthesized version of the unmasked portion of the masked input medical image and synthesized patterns in the masked portion of the masked input medical image. The synthesized patterns is fused with the input medical image to generate a final synthesized medical image.

First claim

The invention claimed is:

1. A computer implemented method comprising:
receiving an input medical image;
generating a synthesized segmentation mask by:
sampling locations from a spatial probability map of abnormality patterns of a disease,
mapping the sampled locations from the spatial probability map to an image space of the synthesized segmentation mask,
generating individual masks each corresponding to a connected component region and positioned at a respective location of the mapped sampled locations in the image space of the synthesized segmentation mask, and
combining the individual masks to generate the synthesized segmentation mask;

masking the input medical image based on the synthesized segmentation mask, the masked input medical image having an unmasked portion and a masked portion;
generating an initial synthesized medical image using a trained machine learning based generator network, the initial synthesized medical image comprising a synthesized version of the unmasked portion of the masked input medical image and synthesized abnormality patterns of the disease in the masked portion of the masked input medical image;
blending the initial synthesized medical image with the input medical image to generate a blended image; and
fusing the synthesized abnormality patterns extracted from the blended image with the input medical image to generate a final synthesized medical image.

2. The computer implemented method of claim 1 , wherein the disease is COVID-19 (coronavirus disease 2019) and the synthesized abnormality patterns comprise one or more of ground glass opacities (GGO), consolidation, and crazy-paving pattern.

3. The computer implemented method of claim 1 , wherein the disease is at least one of a viral pneumonia, a bacterial pneumonia, a fungal pneumonia, and a mycoplasma pneumonia.

4. The computer implemented method of claim 1 , wherein generating individual masks each corresponding to a connected component region and positioned at a respective location of the mapped sampled locations in the image space of the synthesized segmentation mask comprises:
for each of the individual masks:
selecting a number of points on a surface of a mesh of a sphere; and
applying a transformation to each particular point, wherein the transformation applied to a particular point is propagated to neighboring vertices on the surface of the mesh based on a distance between the particular point and each of the neighboring vertices as compared to a distance threshold.

5. The computer implemented method of claim 1 , wherein fusing the synthesized abnormality patterns extracted from the blended image with the input medical image to generate a final synthesized medical image comprises:
smoothing boundaries of the synthesized segmentation mask to generate a smooth synthesized segmentation mask;
cropping masked portions of the smooth synthesized segmentation mask from the blended image to extract the synthesized abnormality patterns;
cropping unmasked portions of the smooth synthesized segmentation mask from the input medical image to extract remaining regions of the input medical image; and
combining the extracted synthesized abnormality patterns and the extracted remaining regions.

6. The computer implemented method of claim 1 , further comprising:
training a machine learning based system for performing a medical image analysis task based on the final synthesized medical image.

7. An apparatus comprising:
means for receiving an input medical image;
means for generating a synthesized segmentation mask by:
means for sampling locations from a spatial probability map of abnormality patterns of a disease,
means for mapping the sampled locations from the spatial probability map to an image space of the synthesized segmentation mask,
means for generating individual masks each corresponding to a connected component region and positioned at a respective location of the mapped sampled locations in the image space of the synthesized segmentation mask, and
means for combining the individual masks to generate the synthesized segmentation mask;

means for masking the input medical image based on the synthesized segmentation mask, the masked input medical image having an unmasked portion and a masked portion;
means for generating an initial synthesized medical image using a trained machine learning based generator network, the initial synthesized medical image comprising a synthesized version of the unmasked portion of the masked input medical image and synthesized abnormality patterns of the disease in the masked portion of the masked input medical image;
means for blending the initial synthesized medical image with the input medical image to generate a blended image; and
means for fusing the synthesized abnormality patterns extracted from the blended image with the input medical image to generate a final synthesized medical image.

8. The apparatus of claim 7 , wherein the disease is COVID-19 (coronavirus disease 2019) and the synthesized abnormality patterns comprise one or more of ground glass opacities (GGO), consolidation, and crazy-paving pattern.

9. The apparatus of claim 7 , wherein the means for generating individual masks each corresponding to a connected component region and positioned at a respective location of the mapped sampled locations in the image space of the synthesized segmentation mask comprises:
means for selecting, for each of the individual masks, a number of points on a surface of a mesh of a sphere; and
means for applying, for each of the individual masks, a transformation to each particular point, wherein the transformation applied to a particular point is propagated to neighboring vertices on the surface of the mesh based on a distance between the particular point and each of the neighboring vertices as compared to a distance threshold.

10. A non-transitory computer readable medium storing computer program instructions, the computer program instructions when executed by a processor cause the processor to perform operations comprising:
receiving an input medical image;
generating a synthesized segmentation mask by:
sampling locations from a spatial probability map of abnormality patterns of a disease,
mapping the sampled locations from the spatial probability map to an image space of the synthesized segmentation mask.
generating individual masks each corresponding to a connected component region and positioned at a respective location of the mapped sampled locations in the image space of the synthesized segmentation mask, and
combining the individual masks to generate the synthesized segmentation mask;

masking the input medical image based on the synthesized segmentation mask, the masked input medical image having an unmasked portion and a masked portion;
generating an initial synthesized medical image using a trained machine learning based generator network, the initial synthesized medical image comprising a synthesized version of the unmasked portion of the masked input medical image and synthesized abnormality patterns of the disease in the masked portion of the masked input medical image;
blending the initial synthesized medical image with the input medical image to generate a blended image; and
fusing the synthesized abnormality patterns extracted from the blended image with the input medical image to generate a final synthesized medical image.

11. The non-transitory computer readable medium of claim 10 , wherein the disease is COVID-19 (coronavirus disease 2019) and the synthesized abnormality patterns comprise one or more of ground glass opacities (GGO), consolidation, and crazy-paving pattern.

12. The non-transit

Assignees

• Siemens Healthcare Gmbh

Inventors

• Liu Siqi
• Georgescu Bogdan
• Xu Zhoubing
• Yoo Youngjin
• Chabin Guillaume
• Chaganti Shikha
• Grbic Sasa
• Piat Sebastien
• Teixeira Brian
• Re Thomas
• Comaniciu Dorin

Classifications

• G06T7/0012 Primary

  Biomedical image inspection · CPC title

• G06T7/11

  Region-based segmentation · CPC title

• G06T17/205

  Re-meshing · CPC title

• G06T2207/10081

  Computed x-ray tomography [CT] · CPC title

• G06T2207/20081

  Training; Learning · CPC title