Systems and methods for segmenting a medical image sequence

Patent metadata

Field	Value
Publication number	US-12482108-B2
Application number	US-202318209704-A
Country	US
Kind code	B2
Filing date	Jun 14, 2023
Priority date	Jun 14, 2023
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

An apparatus may obtain a sequence of medical images of a target structure and determine, using a first ANN, a first segmentation and a second segmentation of the target structure based on a first medical image and a second medical image, respectively. The first segmentation may indicate a first plurality of pixels that may belong to the target structure. The second segmentation may indicate a second plurality of pixels that may belong to the target structure. The apparatus may identify, using a second ANN, a first subset of true positive pixels among the first plurality of pixels that may belong to the target structure, and a second subset of true positive pixels among the second plurality of pixels that may belong to the target structure. The apparatus may determine a first refined segmentation and a second refined segmentation of the target structure based on the true positive pixels.

First claim

What is claimed is:

1 . An apparatus, comprising:
a processor configured to:
obtain a sequence of medical scan images associated with a target structure, wherein the sequence of medical scan images includes a first medical scan image and a second medical scan image;
determine, using a first artificial neural network (ANN), a first segmentation of the target structure and a second segmentation of the target structure based on the first medical scan image and the second medical scan image, respectively, wherein the first segmentation indicates a first plurality of pixels of the first medical scan image predicted by the first ANN as belonging to the target structure, and wherein the second segmentation indicates a second plurality of pixels of the second medical scan image predicted by the first ANN as belonging to the target structure;
extract, using a second ANN, a first set of features associated with the first plurality of pixels and a second set of features associated with the second plurality of pixels;
identify a first subset of true positive pixels among the first plurality of pixels and a second subset of true positive pixels among the second plurality of pixels as truly belonging to the target structure, wherein the identification is made based on a determination that a portion of the first set of features that corresponds to the first subset of true positive pixels is substantially similar to a portion of the second set of features that corresponds to the second subset of true positive pixels; and
determine a first refined segmentation of the target structure and a second refined segmentation of the target structure based on the first subset of true positive pixels and the second subset of true positive pixels, respectively.

2 . The apparatus of claim 1 , wherein the processor is further configured to identify a first subset of false positive pixels and a second subset of false positive pixels from the first plurality of pixels and the second plurality of pixels, respectively, the first subset of false positive pixels comprising pixels from the first medical scan image that have been falsely predicted by the first ANN as belonging to the target structure, the second subset of false positive pixels comprising pixels from the second medical scan image that have been falsely predicted by the first ANN as belonging to the target structure.

3 . The apparatus of claim 2 , wherein the processor being configured to identify the first subset of false positive pixels and the second subset of false positive pixels comprises the processor being configured to determine that the portion of the first set of features that corresponds to the first subset of true positive pixels is substantially different than a portion of the first set of features that corresponds to the first subset of false positive pixels or a portion of the second set of features that corresponds to the second subset of false positive pixels.

4 . The apparatus of claim 3 , wherein the second ANN is trained using a contrastive learning technique to minimize a difference between the respective features associated with the first subset of true positive pixels and the second subset of true positive pixels, and to maximize at least one of a difference between the respective features associated with the first subset of true positive pixels and the first subset of false positive pixels, or a difference between the respective features associated with the first subset of true positive pixels and the second subset of false positive pixels.

5 . The apparatus of claim 2 , wherein the processor being configured to identify the first subset of true positive pixels and the second subset of true positive pixels comprises the processor being configured to:
identify, from the first set of features, one or more features associated with a false positive pixel of the first medical scan image; identify, from the second set of features, one or more features associated with a pixel of the second medical scan image; and determine whether the pixel of the second medical scan image belongs to the target structure based on a weighted sum of features calculated from the one or more features identified from the first set of features and the one or more features identified from the second set of features.

6 . The apparatus of claim 1 , wherein the second ANN is characterized by a recurrent structure via which a first feature extracted by the second ANN from the first medical scan image is used as an input to determine a second feature associated with the second medical scan image.

7 . The apparatus of claim 6 , wherein the recurrent structure includes one or more encoder modules and one or more decoder modules.

8 . The apparatus of claim 6 , wherein the second ANN is trained using at least a first training image and a second training image, and wherein, during the training of the second ANN, features extracted from the first training image are used to determine features associated with the second training image and treated as constants during backpropagation.

9 . The apparatus of claim 1 , wherein the target structure includes a blood vessel, a catheter, or a guide wire, the first medical scan image includes a first x-ray fluoroscopy image, and the second medical scan image includes a second x-ray fluoroscopy image.

10 . A method for processing medical scan images, the method comprising:
obtaining a sequence of medical scan images associated with a target structure, wherein the sequence of medical scan images includes a first medical scan image and a second medical scan image; determining, using a first artificial neural network (ANN), a first segmentation of the target structure and a second segmentation of the target structure based on the first medical scan image and the second medical scan image, respectively, wherein the first segmentation indicates a first plurality of pixels of the first medical scan image predicted by the first ANN as belonging to the target structure, and wherein the second segmentation indicates a second plurality of pixels of the second medical scan image predicted by the first ANN as belonging to the target structure; extracting, using a second ANN, a first set of features associated with the first plurality of pixels and a second set of features associated with the second plurality of pixels; identifying a first subset of true positive pixels among the first plurality of pixels and a second subset of true positive pixels among the second plurality of pixels as truly belonging to the target structure, wherein the identification is made based on a determination that a portion of the first set of features that corresponds to the first subset of true positive pixels is substantially similar to a portion of the second set of features that corresponds to the second subset of true positive pixels; and determining a first refined segmentation of the target structure and a second refined segmentation of the target structure based on the first subset of true positive pixels and the second subset of true positive pixels, respectively.

11 . The method of claim 10 , further comprising identifying a first subset of false positive pixels and a second subset of false positive pixels from the first plurality of pixels and the second plurality of pixels, respectively, the first subset of false positive pixels comprising pixels from the first medical scan image that have been falsely predicted by the first ANN as belonging to the target structure, the second subset of false positive pixels comprising pixels from the second medical scan image that have been falsely predicted by the first ANN as belonging to the target structure.

Assignees

• Shanghai United Imaging Intelligence Co Ltd

Inventors

• Liu Yikang
• Chen Zhang
• Chen Xiao
• Sun Shanhui
• Chen Terrence

Classifications

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/10121

  Fluoroscopy · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/30101

  Blood vessel; Artery; Vein; Vascular · CPC title

• G06T2207/10016

  Video; Image sequence · CPC title