Automated organ risk segmentation machine learning methods and systems
Certain examples provide an image data processing system including an anatomy detector to detect an anatomy in an image and to remove items not included in the anatomy from the image. The example system includes a bounding box generator to generate a bounding box around a region of interest in the anatomy. The example system includes a voxel-level segmenter to classify image data within the bounding box at the voxel level to identify an object in the region of interest. The example system includes an output imager to output an indication of the object identified in the region of interest segmented in the image.
1. An image data processing system comprising:
an anatomy detector to detect an anatomy in an image and to remove items not included in the anatomy from the image using a first machine learning model to identify a region of interest in the anatomy or including the anatomy;
a bounding box generator to generate a bounding box around the region of interest using a second machine learning model and to provide image data within the bounding box;
a voxel-level segmenter to classify the image data within the bounding box at the voxel level using at least two models including a third machine learning model to identify an object in the region of interest;
a model stacker to combine a plurality of models including the at least two models of the voxel-level segmenter to, using image intensity values, a fully connected neural network, and a convolutional neural network, identify image features and produce a fully connected neural network model and a gradient boosting machine to classify the object in the region of interest and determine at least one feature related to the object; and
an output imager to output an indication of the object identified in the region of interest segmented in the image, the fully connected neural network model, and the gradient boosting machine.
2. The system of claim 1 , wherein the third machine learning model includes at least one of a first deep learning model or a second deep learning model, and wherein the at least two models employed by the voxel-level segmenter include the first deep learning model to classify a target voxel and neighboring voxels and the second deep learning model to classify the target voxel and neighboring patches.
3. The system of claim 1 , wherein the object includes an organ.
4. The system of claim 1 , wherein the image includes a computed tomography image.
5. The system of claim 4 , wherein the computed tomography image is a three-dimensional computed tomography image processed slice-by-slice to identify the object.
6. The system of claim 1 , wherein the bounding box generator is to generate a first bounding box in an axial image slice, a second bounding box in a coronal image slice, and a third bounding box in a sagittal image slice and to generate the bounding box around the region of interest in a three-dimensional image by combining the first bounding box, the second bounding box, and the third bounding box.
7. The system of claim 6 , wherein the bounding box generator is to be trained using a deep learning network and a database of positive and negative classification results including at least one rotated image slice, at least one translated image slice, and at least one duplicated image slice.
8. A non-transitory computer-readable storage medium including instructions which, when executed, cause a processor to at least implement a method of image processing comprising:
detecting an anatomy in an image using a first machine learning model to identify a region of interest in the anatomy or including the anatomy;
removing items not included in the anatomy from the image;
generating a bounding box around the region of interest using a second machine learning model and providing image data within the bounding box;
classifying image data within the bounding box at the voxel level using at least two models including a third machine learning model to identify an object in the region of interest;
combining a plurality of models including the at least two models to, using image intensity values, a fully connected neural network, and a convolutional neural network, identify image features and produce a fully connected neural network model and a gradient boosting machine to classify the object in the region of interest and determine at least one feature related to the object; and
outputting an indication of the object identified in the region of interest segmented in the image, the fully connected neural network model, and the gradient boosting machine.
9. The computer-readable storage medium of claim 8 , wherein the third machine learning model includes at least one of a first deep learning model or a second deep learning model, and wherein the at least two models include the first deep learning model to classify a target voxel and neighboring voxels and the second deep learning model to classify the target voxel and neighboring patches.
10. The computer-readable storage medium of claim 8 , wherein the object includes an organ.
11. The computer-readable storage medium of claim 8 , wherein the image includes a computed tomography image.
12. The computer-readable storage medium of claim 11 , wherein the computed tomography image is a three-dimensional computed tomography image processed slice-by-slice to identify the object.
13. The computer-readable storage medium of claim 8 , wherein generating a bounding box around a region of interest in the anatomy further includes:
generating a first bounding box in an axial image slice, a second bounding box in a coronal image slice, and a third bounding box in a sagittal image slice; and
generating the bounding box around the region of interest in a three-dimensional image by combining the first bounding box, the second bounding box, and the third bounding box.
14. The computer-readable storage medium of claim 13 , further including training at least one model using a deep learning network and a database of positive and negative classification results including at least one rotated image slice, at least one translated image slice, and at least one duplicated image slice.
15. A computer-implemented method of image processing comprising:
detecting, using at least one processor, an anatomy in an image using a first machine learning model to identify a region of interest in the anatomy or including the anatomy;
removing, using the at least one processor, items not included in the anatomy from the image;
generating, using the at least one processor, a bounding box around the region of interest using a second machine learning model and providing image data within the bounding box;
classifying, using the at least one processor, the image data within the bounding box at the voxel level using at least two models including a third machine learning model to identify an object in the region of interest;
combining, using the at least one processor, a plurality of models including the at least two models to, using image intensity values, a fully connected neural network, and a convolutional neural network, identify image features and produce a fully connected neural network model and a gradient boosting machine to classify the object in the region of interest and determine at least one feature related to the object; and
outputting, using the at least one processor, an indication of the object identified in the region of interest segmented in the image, the fully connected neural network model, and the gradient boosting machine.