Systems and methods for image cropping and anatomical structure segmentation in medical imaging
AssigneeVarian Medical Systems International AG
InventorsHannu Laaksonen, Janne Nord, Jan Schreier
Life of this patent
Estimating the term…
Abstract
One or more medical images of a patient are processed by a first neural network model to determine a region-of-interest (ROI) or a cut-off plane. Information from the first neural network model is used to crop the medical images, which serves as input to a second neural network model. The second neural network model processes the cropped medical images to determine contours of anatomical structures in the medical images of the patient. Each of the first and second neural network models are deep neural network models. By use of cropped images in the training and inference phases of the second neural network model, contours are produced with sharp edges or flat surfaces.
Claims35
A method comprising:
- training at least one first neural network model using a first data set of medical images, each of the medical images in the first data set including a contour of an anatomical structure therein and at least one of a region-of-interest (ROI) around the anatomical structure and a cut-off plane with respect to the anatomical structure, each first neural network model being trained to approximate the ROI or the cut-off plane;
- based on data from the at least one first neural network model, cropping medical images of a second data set, each of the medical images in the second data set including a contour of the anatomical structure;
- and
- training at least one second neural network model using the cropped medical images, each second model being trained to approximate the contour of the anatomical structure.
A system comprising:
- one or more data storage devices storing at least one first neural network model and at least one second neural network model, each first neural network model having been trained to approximate a region-of-interest (ROI) around an anatomical structure or a cut-off plane with respect to the anatomical structure, each second neural network model having been trained to approximate a contour of the anatomical structure;
- and
- one or more processors operatively coupled to the one or more data storage devices and configured to:
- employ the at least one first neural network model to process one or more medical images of a patient to generate cropping data;
- based at least one the cropping data from the at least one first neural network model, crop the one or more medical images of the patient;
- and
- employ the at least one second neural network model to process the one or more cropped medical images of the patient to generate one or more contours of anatomical structures in the medical images of the patient.
A non-transitory computer-readable storage medium upon which is embodied a sequence of programmed instructions, and a computer processing system that executes the sequence of programmed instructions embodied on the computer-readable storage medium to cause the computer processing system to:
- process one or more medical images of a patient using at least one first neural network model to generate cropping data, each first neural network model having been trained to approximate a region-of-interest (ROI) around an anatomical structure or a cut-off plane with respect to the anatomical structure;
- crop the one or more medical images of the patient based at least on the cropping data from the at least one first neural network model;
- and
- process the one or more cropped medical images of the patient using at least one second neural network model to generate one or more contours of anatomical structures in the one or more medical images of the patient, each second neural network model having been trained to approximate contours of anatomical structures.
Dependencies are inferred from the claim text.