Simulating pathology images based on anatomy data
Systems and methods are provided for an image processing system. In an example, a method includes acquiring a pathology dataset, acquiring a reference dataset, generating a deformation field by mapping points of a reference case of the reference dataset to points of a patient image of the pathology dataset, manipulating the deformation field, applying the deformation field to the reference case to generate a simulated pathology image including a simulated deformation pathology, and outputting the simulated pathology image.
1 . A method, comprising:
acquiring a pathology dataset;
acquiring a reference dataset;
generating a deformation field by mapping points of a reference case of the reference dataset to points of a patient image of the pathology dataset;
manipulating the deformation field by adjusting parameters of a deformation algorithm based on additional information including one or more of tissue elasticity parameters, prior knowledge of deformities, and deformation parameters of different tissue types to adjust regions of deformation within the deformation field;
applying the deformation field to the reference case to generate a simulated pathology image including a simulated deformation pathology; and
outputting the simulated pathology image.
2 . The method of claim 1 , wherein the pathology dataset includes a plurality of patient images, each including an anatomical region having a deformation pathology.
3 . The method of claim 2 , wherein generating the deformation field includes non-rigidly registering the reference case to the patient image.
4 . The method of claim 2 , wherein generating the deformation field includes group-wise non-rigid registration of the plurality of patient images of the pathology dataset to the reference case of the reference dataset.
5 . The method of claim 1 , wherein the pathology dataset is anatomical knowledge of an anatomy, including characteristics and parameters of the anatomy which differentiate the anatomy having disease pathology from the anatomy without disease pathology.
6 . The method of claim 5 , wherein generating the deformation field includes performing anatomy-aware deformation of the reference case based on the patient image.
7 . The method of claim 6 , wherein performing anatomy-aware deformation includes applying an anatomy mask to the reference case, where the anatomy mask indicates a region of interest to be deformed.
8 . The method of claim 1 , wherein the reference case is an image of an anatomical region without disease pathology or an image of the anatomical region with disease pathology.
9 . The method of claim 1 , wherein manipulation of the deformation field by adjusting parameters of a deformation algorithm based on the additional information includes at least one of adjusting a degrees of freedom parameter of the deformation field or adjusting degrees of deformation according to tissue-type parameters, and wherein the additional information includes tissue elasticity parameters, prior knowledge of deformities, and deformation parameters of different tissue types to adjust regions of deformation within the deformation field.
10 . The method of claim 1 , further comprising manipulating the simulated pathology image by adjusting image intensities to simulate artifacts such as metal implants, calcification, and lesions.
11 . The method of claim 10 , wherein manipulating the simulated pathology image includes inputting the simulated pathology image into a machine learning model trained to output a realistic depiction of anatomy simulated in the simulated pathology image.
12 . The method of claim 1 , wherein outputting the simulated pathology image includes outputting the simulated pathology image to a memory which is communicably coupled to a training module including an untrained machine learning model and instructions for training the untrained machine learning model.
13 . A method, comprising:
generating a deformation field based on a pathology dataset and a reference case, wherein generating the deformation field comprises performing non-rigid registration of the reference case for a first evolution period using a deformation algorithm;
manipulating the deformation field;
applying the manipulated deformation field to the reference case to generate a simulated pathology image;
augmenting the simulated pathology image; and
outputting the simulated pathology image.
14 . The method of claim 13 , wherein the pathology dataset is a medical image of an embodiment of an anatomy with at least one anatomical deformity, or parameters and characteristics of at least one anatomical deformity of the anatomy.
15 . The method of claim 13 , wherein the reference case is an embodiment of an anatomy of the pathology dataset without anatomical deformity or with anatomical deformity.
16 . The method of claim 13 , wherein the deformation algorithm is an ANTS algorithm, and wherein the ANTS algorithm has a first symmetric normalization value and a first Gaussian displacement value to control smoothing.
17 . The method of claim 13 , wherein augmenting the simulated pathology image includes modulating intensities of the simulated pathology image due to a bio-physical tissue model.
18 . An image processing system, comprising:
a user input device;
a display device;
a memory, wherein the memory includes machine readable instructions;
a training module, wherein the training module includes an untrained machine learning model; and
a processor, wherein the processor is communicably coupled to the user input device, the display device, the training module, and the memory, and when executing the machine readable instructions, the processor is configured to:
acquire a pathology dataset;
acquire a reference dataset;
generate a deformation field by mapping points of a reference case of the reference dataset to points of a patient image of the pathology dataset, wherein the deformation field comprises performing non-rigid registration of the reference case for a controlled evolution period using a deformation algorithm with smoothing parameters to control deformation;
manipulate the deformation field;
apply the deformation field to the reference case to generate a simulated pathology image; and
output the simulated pathology image to the memory.
19 . The image processing system of claim 18 , wherein a pathology type of the simulated pathology image is labeled, and a training pair is formed of at least one pathology label coupled to the simulated pathology image.
20 . The image processing system of claim 19 , wherein, when executing the machine readable instructions, the processor is further configured to:
acquire a plurality of training pairs from the memory;
input the simulated pathology image of the training pair into the untrained machine learning model;
extract features from the simulated pathology image using the untrained machine learning model;
predict at least one predicted pathology label for the simulated pathology image based on extracted features;
compare the at least one predicted pathology label to the at least one pathology label coupled to the simulated pathology image and calculate loss;
adjust weights and biases of the untrained machine learning model based on loss to train the untrained machine learning model; and
output a machine learning model trained to identify pathologies in medical images captured by an imaging system, wherein the imaging system is communicably coupled to the training module.