Tumor tissue characterization using multi-parametric magnetic resonance imaging
Brain tumor or other tissue classification and/or segmentation is provided based on from multi-parametric MRI. MRI spectroscopy, such as in combination with structural and/or diffusion MRI measurements, are used to classify. A machine-learned model or classifier distinguishes between the types of tissue in response to input of the multi-parametric MRI. To deal with limited training data for tumors, a patch-based system may be used. To better assist physicians in interpreting results, a confidence map may be generated using the machine-learned classifier.
1. A method for characterizing tumor tissue from multi-parametric imaging by a magnetic resonance (MR) scanner, the method comprising:
scanning a patient by the MR scanner, the scanning including structural, diffusion, and spectroscopy measurements of the patient;
classifying the tumor tissue of the patient between two or more classes of tumor type, the classifying by a machine-learned neural network in response to input of the structural, diffusion, and spectroscopy measurements;
generating a confidence map comprising confidence values for each voxel that indicate a consistency of the machine-learned neural network's classification predictions over different network configurations; and
displaying an image of the tumor tissue segmented based on the two or more classes and the confidence map.
2. The method of claim 1 wherein scanning comprises scanning with the structural measurements including two or more of T1, T2, FLAIR, and T1 Post-Contrast measurements.
3. The method of claim 1 wherein scanning comprises scanning with the diffusion measurements including two or more of diffusion tensor imaging (DTI) axial diffusivity, DTI fractional anisotropy, DTI mean diffusivity, and diffusion weighted imaging (DWI) B0 measurements.
4. The method of claim 1 wherein scanning comprises scanning with the spectroscopy measurements including two or more of N-acetyl aspartate, creatine, lactate, water, gix, inositol, and choline measurements.
5. The method of claim 1 wherein the machine-learned neural network comprises a fully connected neural network.
6. The method of claim 1 wherein classifying comprises classifying by application of the machine-learned neural network patch-by-patch for different locations from the measurements.
7. The method of claim 1 wherein classifying comprises classifying as enhanced, edema, necrosis, and cavity tumor types.
8. The method of claim 1 wherein classifying comprises classifying between the two or more classes of tumor type and a class of normal tissue.
9. The method of claim 1 wherein classifying comprises classifying by the machine-learned neural network in response to input of the structural, diffusion, and spectroscopy measurements and a normal tissue mask.
10. The method of claim 1 further comprising co-registering the structural, diffusion, and spectroscopy measurements of the patient prior to the input to the machine-learned neural network.
11. The method of claim 1 further comprising locating the tumor tissue, wherein classifying comprises classifying the tumor tissue and tissue adjacent the tumor tissue without classifying tissue located a predetermined number of patches from the tissue adjacent the tumor tissue.
12. The method of claim 1 wherein scanning comprises scanning a brain of the patient, and wherein classifying comprises classifying the tumor tissue as brain tumor tissue.
13. The method of claim 1 wherein the machine-learned neural network was trained on training data of patches using weighted sampling of normal tissue related to tumor tissue.
14. The method of claim 1 , wherein for the different network configurations half of the nodes in a hidden layer and their corresponding connections to other nodes are randomly removed from the machine-learned neural network.