Systems and methods for image cropping and anatomical structure segmentation in medical imaging
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.
1. A method of generating a contour of an anatomical structure on a data set of medical images, comprising:
processing the data set of medical images to obtain a data set of cropped medical images;
processing the data set of cropped medical images by a plurality of trained segmentation neural network models working separately or in unison to obtain respective contour data outputs, each of the trained segmentation neural network models having been trained to approximate the contour of the anatomical structure; and
combining the obtained contour data outputs to obtain a singular output of contour data,
wherein the processing of the data set of medical images includes:
processing the medical images using one or more trained neural network models to generate corresponding crop data outputs, the neural network models having been trained to approximate a cut-off plane with respect to the anatomical structure; and
cropping the medical images based on the generated crop data outputs,
the method further comprising:
generating a contour using the singular output of contour data; and
modifying the generated contour to have a substantially straight edge by applying an algorithm that enforces straight cut-offs.
2. The method of claim 1 , further comprising combining the singular output of contour data with the medical images to obtain an overlay of the contour with the anatomical structure on the medical images.
3. The method of claim 1 , wherein the singular output of contour data is an overlay of the contour with the anatomical structure on the medical images.
4. The method of claim 1 , wherein the trained segmentation neural network models have different operating characteristics, including different input image resolutions, different hidden layer configurations, and different underlying neural network models.
5. The method of claim 1 , wherein the cropping is with respect to images input to each of the plurality of trained segmentation neural network models.
6. The method of claim 1 , wherein the cropping is with respect to images input to only one of the trained segmentation neural network models.
7. The method of claim 1 , wherein the cropping is with respect to images input to a subset of the trained segmentation neural network models.
8. The method of claim 1 , further comprising combining the crop data outputs from the trained neural network models to generate a single output of cropping instructions.
9. The method of claim 1 , wherein the crop data outputs from the trained neural network models are used separately to crop images input to specific trained segmentation neural network models.
10. A system for generating a contour of an anatomical structure on a data set of medical images, comprising:
one or more data storage devices storing one or more neural network models which have been trained to approximate a cut-off plane with respect to the anatomical structure, and a plurality of segmentation neural network models which have been trained to approximate the contour of the anatomical structure; and
one or more processors operatively coupled to the one or more data storage devices and configured to:
process the data set of medical images using the one or more trained neural network models to generate corresponding crop data outputs;
crop the medical images based on the generated crop data outputs to obtain cropped medical images;
process the cropped medical images using the plurality of trained segmentation neural network models working separately or in unison to obtain respective contour data outputs; and
combine the contour data outputs to obtain a singular output of contour data,
wherein the one or more processors are further configured to:
generate a contour using the singular output of contour data; and
modify the generated contour to have a substantially straight edge by applying an algorithm that enforces straight cut-offs.
11. The system of claim 10 , wherein the one or more processors are further configured to combine the singular output of contour data with the medical images to obtain an overlay of the contour with the anatomical structure on the medical images.
12. The system of claim 10 , wherein the trained segmentation neural network models have different operating characteristics.
13. The system of claim 10 , wherein the cropping is with respect to images input to each of the plurality of trained segmentation neural network models.
14. The system of claim 10 , wherein the cropping is with respect to images input to only one of the trained segmentation neural network models.
15. The system of claim 10 , wherein the cropping is with respect to images input to a subset of the trained segmentation neural network models.
16. The system of claim 10 , wherein the one or more processors are further configured to combine the crop data outputs from the trained neural network models to generate a single output of cropping instructions.
17. The system of claim 10 , wherein the one or more processors are further configured to use the crop data outputs from the trained neural network models separately to crop images input to specific trained segmentation neural network models.