Systems, devices, and methods for segmentation of anatomical image data
Systems, devices, and methods for segmentation of patient anatomy are described herein. A method can include receiving a three-dimensional (3D) scan volume including a set of images of a 3D region of patient anatomy. The 3D region of patient anatomy can include a set of anatomical structures. The method can also include generating a set of two-dimensional (2D) radiographs using the 3D scan volume. Each 2D radiograph from the set of 2D radiographs can include 3D image data extracted from the 3D scan volume. The method can also include training a segmentation model to segment 2D radiographs using the set of 2D radiographs to identify one or more anatomical parts of interest.
1 . A method, comprising:
receiving a three-dimensional (3D) scan volume including a set of Magnetic Resonance Imaging (MRI) images of a 3D region of patient anatomy, the 3D region of patient anatomy including a set of anatomical structures, the MRI images including images in a first anatomical plane and images in a second anatomical plane different from the first anatomical plane;
processing the set of MRI images using a segmentation model in which the segmentation model receives the set of MRI images as inputs and processes the images in the first and second anatomical planes;
generating a segmentation output in response to processing the set of MRI images using the segmentation model;
combining the three-dimensional scan volume with the segmentation output to generate information of higher dimensionality of the set of anatomical structures; and
generating a 3D anatomical model that identifies one or more anatomical parts of interest in the 3D scan volume based on the segmentation output and the information of higher dimensionality.
2 . The method of claim 1 , wherein the first anatomical plane is a sagittal plane and the second anatomical plane is an axial plane.
3 . The method of claim 1 , further comprising visualizing the segmentation output.
4 . The method of claim 1 , wherein the one or more anatomical parts of interest include an intervertebral disc.
5 . The method of claim 1 , wherein the segmentation model is trained using a training dataset, the training dataset including MRI images in the first anatomical plane and MRI images in the second anatomical plane.
6 . The method of claim 1 , wherein the segmentation model includes a convolutional neural network (CNN).
7 . The method of claim 1 , the method further comprising prior to processing the set of MRI images, denoising the set of MRI images using the segmentation model.
8 . The method of claim 1 , wherein generating the segmentation output includes: for at least one pixel in an MRI image in the set of MRI images:
determining a probability that the at least one pixel belongs to a first class of a plurality of classes; and
classifying the at least one pixel into the first class based on the probability.
9 . The method of claim 1 , further comprising visualizing the 3D anatomical model.
10 . The method of claim 1 , wherein generating the 3D anatomical model that identifies one or more anatomical parts of interest comprises assigning a label characterizing one or more of the anatomical parts of interest.
11 . An apparatus, comprising:
a computer-readable memory; and
a processor operatively coupled to the computer-readable memory storing instructions, that when executed, cause the processor to:
receive a three-dimensional (3D) scan volume including a set of Magnetic Resonance Imaging (MRI) images of a 3D region of patient anatomy, the 3D region of patient anatomy including a set of anatomical structures, the MRI images including images in a first anatomical plane and images in a second anatomical plane different from the first anatomical plane;
process the set of MRI images using a segmentation model in which the segmentation model receives the set of MRI images as inputs and processes the images in the first and second anatomical planes;
generate a segmentation output in response to processing the set of MRI images using the segmentation model;
combine the three-dimensional scan volume with the segmentation output to generate information of higher dimensionality of the set of anatomical structures; and
generate a 3D anatomical model that identifies one or more anatomical parts of interest in the 3D scan volume based on the segmentation output and the information of higher dimensionality.
12 . The apparatus of claim 11 , wherein the first anatomical plane is a sagittal plane and the second anatomical plane is an axial plane.
13 . The apparatus of claim 11 , further comprising visualizing the segmentation output.
14 . The apparatus of claim 11 , wherein the one or more anatomical parts of interest include an intervertebral disc.
15 . The apparatus of claim 11 , wherein the segmentation model is trained using a training dataset, the training dataset including MRI images in the first anatomical plane and MRI images in the second anatomical plane.
16 . The apparatus of claim 11 , wherein the segmentation model includes a convolutional neural network (CNN).
17 . The apparatus of claim 11 , the processor is further configured to prior to processing the set of MRI images, denoise the set of MRI images using the segmentation model.
18 . The apparatus of claim 11 , wherein the processor is further configured to:
for at least one pixel in an MRI image in the set of MRI images:
determine a probability that the at least one pixel belongs to a first class of a plurality of classes; and
classify the at least one pixel into the first class based on the probability.
19 . The apparatus of claim 11 , wherein generating the 3D anatomical model that identifies one or more anatomical parts of interest comprises assigning a label characterizing one or more of the anatomical parts of interest.
20 . The apparatus of claim 11 , wherein the processor is further configured to visualize the 3D anatomical model.