Characterizing intra-site tumor heterogeneity
A method and a system for measuring intra-site heterogeneity in a tumor using magnetic resonance imaging (MRI). The method includes acquiring magnetic resonance (MR) images using MRI modality; segmenting tumor sites in the MR images; dividing each of the tumor sites into a plurality of sub-regions; deriving image biomarkers from each voxel or pixel in the plurality of sub-regions; classifying each voxel or pixel in the plurality of sub-regions into genotypes or molecular subtypes based on the extracted image biomarkers and a classifier model including associations between image biomarkers and genotypes or molecule subtypes; creating a distribution of genotypes or molecular subtypes in the each of the plurality of sub-regions based on classifications of voxels or pixels; generating spatial information of genotypes or molecular subtypes in the tumor sites based on the distribution; and measuring intra-site heterogeneity in the tumor sites.
1. A method for measuring intra-site heterogeneity in a tumor using magnetic resonance imaging (MRI), the method comprising:
acquiring one or more magnetic resonance (MR) images of a region of interest including the tumor using at least one MRI modality;
segmenting one or more tumor sites in the MR images;
dividing each of the one or more tumor sites into a plurality of sub-regions;
deriving image biomarkers from each voxel or pixel in each of the plurality of sub-regions;
classifying each voxel or pixel in each of the plurality of sub-regions into one of genotypes or molecular subtypes based on the derived image biomarkers and a classifier model including associations between image biomarkers and genotypes or molecule subtypes;
creating a distribution of genotypes or molecular subtypes in the each of the plurality of sub-regions based on classifications of voxels or pixels in each of the plurality of sub-regions;
generating spatial information of genotypes or molecular subtypes in the tumor sites based on the distribution of the genotypes or molecular subtypes in the plurality of sub-regions; and
measuring intra-site heterogeneity in the tumor sites based on the spatial information.
2. The method of claim 1 , wherein the tumor is primary brain tumors or brain metastases.
3. The method of claim 1 , wherein the intra-tumor heterogeneity in the tumor sites comprises different genetic or molecular characterizations within the tumor sites, including one or more of molecules, gene expressions, or mutations.
4. The method of claim 1 , wherein the MR image is acquired by magnetic resonance imaging sequences with or without administration of a contrast agent.
5. The method of claim 4 , wherein magnetic resonance imaging sequences include one of T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging, functional MR imaging, diffusion tensor imaging, susceptibility-weighted imaging, perfusion-weighted imaging, chemical shift imaging, and intra-voxel incoherent motion, and their variations.
6. The method of claim 1 , wherein a spatial resolution of the one or more MR image is less than 2.0 milliliter.
7. The method of claim 1 , wherein a spatial resolution of the one or more MR images is less than 1.0 milliliter.
8. The method of claim 1 , wherein a spatial resolution of the one or more MR images is less than 0.2 milliliter.
9. The method of claim 1 , wherein the sub-region of the tumor site includes at least 100 pixels or voxels.
10. The method of claim 1 , wherein the sub-region of the tumor site includes at least 20 pixels or voxels.
11. The method of claim 1 , wherein the sub-region of the tumor site includes at least 8 pixels or voxels.
12. The method of claim 1 , wherein the image biomarkers include tissue properties or image features derived from the acquired MR image;
wherein the tissue properties include at least one of T1, T2, proton density, perfusion, cerebral blood volume, oxygen concentration, permeability, iron concentration, and diffusion coefficient; and
wherein the image features include at least one of size and shape based-features, descriptors of the image intensity histogram, descriptors of the relationships between image voxels including a gray-level co-occurrence matrix (GLCM), run length matrix (RLM), size zone matrix (SZM), and neighborhood gray tone difference matrix (NGTDM) derived textures, textures extracted from filtered images, and fractal features.
13. The method of claim 1 , wherein a biophysical model is used to associate each of the image biomarkers with a specific genotype or molecule subtype.
14. The method of claim 1 , wherein the classifier model comprises either a supervised model or an unsupervised model;
wherein the supervised model further comprises support vector machines, linear regression, logistic regression, naive Bayes, linear discriminant analysis, decision trees, k-nearest neighbor algorithm, similarity learning and their variations; and
wherein the unsupervised model further comprises Clustering analysis, sample specificity learning, self-supervised learning, and Generative models.
15. The method of claim 1 , wherein the classifier model comprises one or more of machine learning models, including support voting machine (SVM), a naive Bayes classifier, a decision tree, a boosted tree, a random forest classifier, a fuzzy logic classifier, a neural network, a nearest neighbor classifier, deep learning, and a nonlinear classifier.
16. The method of claim 1 , wherein the classifier model is trained by image biomarkers derived from a plurality of training images and ground truth of cancer patients with specific genetic or molecular data; and
wherein each of the plurality of training images corresponds to one or more patients with the tumor.
17. The method of claim 1 , further comprising: determining the genotype of each voxel or pixel in each of the plurality of sub-regions with an area under the curve (AUC) of greater than 0.95.
18. The method of claim 1 , further comprising: determining the genotype of each voxel or pixel in each of the plurality of sub-regions with an area under the curve (AUC) of greater than 0.8.
19. The method of claim 1 , wherein classifying each voxel or pixel in each of the plurality of sub-regions into one of genotypes or molecular subtypes is determined based on a probability of the voxel or pixel being classified into one of genotypes or molecular subtypes.
20. A system for measuring intra-site heterogeneity in a tumor using magnetic resonance imaging (MRI), the system comprising:
a receiver configured to acquire one or more magnetic resonance (MR) images of a region of interest including the tumor using at least one MRI modality; and
a processor configured to:
segment one or more tumor sites in the MR images;
divide each of the one or more tumor sites into a plurality of sub-regions;
derive image biomarkers from each voxel or pixel in each of the plurality of sub-regions;
classify each voxel or pixel in each of the plurality of sub-regions into one of genotypes or molecular subtypes based on the derived image biomarkers and a classifier model including associations between image biomarkers and genotypes or molecule subtypes;
create a distribution of genotypes or molecular subtypes in the each of the plurality of sub-regions based on classifications of voxels or pixels in each of the plurality of sub-regions;
generate spatial information of genotypes or molecular subtypes in the tumor sites based on the distribution of the genotypes or molecular subtypes in the plurality of sub-regions; and
measure intra-site heterogeneity in the tumor sites based on the spatial information.