Automated segmentation using deep learned priors
Embodiments described herein provide a hybrid technique which incorporates learned pulmonary nodule features in a model based energy minimization segmentation using graph cuts. Features are extracted from training samples using a convolutional neural network, and the segmentation cost function is augmented via the deep learned energy. The system and method improves segmentation performance and more robust initialization.
1. A system comprising a hybrid method to process image analytics posed in a cost function minimization framework, the system comprising:
one or more image data sets comprising image-based data that utilizes low level image features; and
a probability map that provides prior knowledge about an object;
wherein the hybrid method comprises: providing at least one of the image-based data, providing a regularizing energy that imparts smoothness by reducing an error at the edges of the probability map, integrating the prior knowledge about the object via the probability map, and extracting features, as specified by an user, for characterization.
2. The system of claim 1 , wherein the low level image features include one or more of edges, texture, and region statistics, alone or in combination.
3. The system of claim 1 , wherein the hybrid method is used as a pre-processing tool by a processor for model initialization, or utilized as a post-processing tool by the processor for refining output data and output images.
4. The system of claim 1 , wherein the hybrid method further comprises a step of predicting the cost function minimization framework.
5. The system of claim 4 , wherein the cost function minimization framework is an optimization architecture that computes an optimal partition to provide visualization of the object delineated from the background.
6. The system of claim 1 , wherein the image analytics comprise segmentation, enhancement, de-noising, and background estimation, individually or in combination.
7. The system of claim 1 , wherein the one or more image data sets are provided by at least one modality, or a combination of modalities, wherein the modalities comprise: magnetic resonance (MR), computed tomography (CT), ultrasound, X-ray, or variations thereof.
8. The system of claim 1 , wherein the hybrid method utilizes a smooth surface of the object to enhance the object in the image data set and suppress background of the image data set.
9. The system of claim 8 , wherein the objects include anatomical structures.
10. The system of claim 9 , wherein the anatomical structures comprise filamentous objects including blood vessels.