System and method for automatic segmentation of tumor sub-compartments in pediatric cancer using multiparametric MRI
AssigneeWisconsin Alumni Research Foundation
InventorsPallavi Tiwari, Rohan Bareja
Life of this patent
Estimating the term…
Abstract
The present disclosure provides systems and methods for accurate segmentation of pediatric tumors using MRI images to improve disease diagnosis, prognosis, patient treatment and surgical planning. Specifically, systems and method herein perform segmentation of pediatric tumor sub-compartments using a fully automated transfer learning approach that learns tumor-specific patterns from adult brain tumors and transfers the knowledge to the pediatric brain tumor domain.
Claims20
A computer system for automatic segmentation of tumor sub-compartments in pediatric magnetic resonance imaging (MRI) data, the computer system comprising:
- a communications connection configured to receive MRI data of a pediatric patient and a parameter from the MRI data to select one of a plurality of age-specific atlases;
- a processor configured to receive the MRI data and the parameter and configured to carry out steps comprising:
- pre-processing the MRI data to generate pre-processed MRI data, wherein the pre-processing includes registering the MRI data to the age-specific atlas;
- segmenting the pre-processed MRI data, wherein segmenting includes inputting the pre-processed MRI data into one or more deep-learning-based models trained on adult MRI data to generate segmentation of tumor sub-compartments in the MRI data of the pediatric patient;
- and
- a display configured to display the segmentation of tumor sub-compartments in the MRI data of the pediatric patient.
A method for automatic segmentation of tumor sub-compartments in pediatric magnetic resonance imaging (MRI) data, the method comprising:
- using a computer processor, access MRI data of a pediatric patient;
- using the computer processor, receive a parameter from the MRI data to select one of a plurality of age-specific atlases;
- using the computer processor, pre-process on the MRI data to generate pre-processed MRI data, wherein the pre-processing includes registering the MRI data to the age-specific atlas;
- and
- using the computer processor, segment the pre-processed MRI data, wherein segmentation includes inputting the pre-processed MRI data into one or more deep-learning-based models and outputting a prediction of an area of one or more tumor sub-compartments in the pre-processed MRI data.
Dependencies are inferred from the claim text.