System and method for attention-based classification of high-resolution microscopy images
This invention provides a system and method for analyzing and classifying imaged from whole slides of tissue. A source of image data transmits images of the tissue on the whole slides to a GPU. The GPU performs a feature extraction process that identifies and segments regions of interests in each of the images, and an attention network that, based upon training from an expert, identifies trained characteristics can comprise cancerous and/or pre-cancerous conditions/e.g. those associated with a gastrointestinal tract, such ad Barret's Esophagus. The feature extraction process can include a convolutional neural network (CNN). The attention network can be adapted performs attention/based weighting of features relative to the trained characteristics, and/or the attention network can include 3D convolutional filters. The image data is acquired using an image sensor having approximately 100 Megapixel resolution.
1. A system for analyzing and classifying images from whole slides of tissue comprising:
a source of image data including images of the tissue on the whole slides, each of the images divided into a plurality of adjacent non-overlapping tiles;
a computer processor executing a CNN-based feature extraction process by which the processor identifies regions of interest in each non-overlapping tile of the images thereby producing a grid-based feature map of the identified regions of interest; and
an attention network that, based upon training from an expert, when executed by the processor, configures the processor to identify trained characteristics in the regions of interest of the grid-based feature map and provide identification data for a least a portion of the identified regions of interest to a user,
wherein the attention network-configured processor performs attention-based weighting of features relative to the trained characteristics, and
wherein the attention network includes 3D convolutional filters of size N×d×d, where N is a depth of a filter kernel and d denotes a height and width of the kernel.
2. The system as set forth in claim 1 wherein the characteristics comprise medical conditions.
3. The system as set forth in claim 2 wherein the medical conditions comprise at least one of cancerous and pre-cancerous conditions.
4. The system as set forth in claim 3 wherein the tissue is associated with a gastrointestinal tract of the patient.
5. The system as set forth in claim 1 wherein the processor includes a GPU that operates the feature extraction process and executes the attention network.
6. The system as set forth in claim 1 wherein the image data of each whole slide is acquired using an image sensor having at least approximately 100 Megapixel resolution.
7. A method for analyzing and classifying images from whole slides of tissue comprising the steps of:
acquiring image data including images of the tissue on the whole slides;
dividing each of the images into a plurality of adjacent non-overlapping tiles;
extracting features with a CNN-configured processor by identifying regions of interest in each non-overlapping tile of the images; and
based upon training from an expert, identifying, with a processor operating an attention network, trained characteristics in one or more of the regions of interest and providing identification data for at least a portion of the regions of interest to a user
wherein the step of the processor operating the attention network comprises performing attention-based weighting of features relative to the trained characteristics, and
wherein the attention network includes 3D convolutional filters of size N×d×d, where N is a depth of a filter kernel and d denotes a height and width of the kernel.
8. The method as set forth in claim 7 wherein the characteristics comprise at least one of visible tissue-related medical conditions, cancerous conditions and pre-cancerous conditions.
9. The method as set forth in claim 8 wherein the tissue is associated with a gastrointestinal tract of the patient.
10. The method as set forth in claim 7 further comprising a GPU that operates the step of extracting and that operates the attention network.
11. A non-transitory, computer-readable medium including program instructions that when executed by a computer processor perform the steps of:
extracting, with a trained CNN, features from acquired image data, including images of the tissue on whole slides divided into a plurality of adjacent non-overlapping tiles, by identifying regions of interest in each of at least a portion of the tiles of the images; and
based upon training from an expert, identifying, by operating an attention network, trained characteristics in the regions of interest and providing identification data for at least a portion of the regions of interest to a user through an interface,
wherein the step of operating the attention network comprises performing attention-based weighting of features relative to the trained characteristics, and
wherein the attention network applies 3D convolutional filters of size N×d×d to the regions of interest, where N is a depth of a filter kernel and d denotes a height and width of the kernel.
12. The non-transitory, computer-readable medium as set forth in claim 11 wherein the characteristics comprise at least one of visible tissue-related medical conditions, cancerous conditions and pre-cancerous conditions.
13. The non-transitory, computer-readable medium as set forth in claim 12 wherein the tissue is associated with a gastrointestinal tract of the patient.
14. The method as set forth in claim 7 wherein the processor includes a GPU that performs the extracting of features and the operating of the attention network.
15. The method as set forth in claim 7 wherein the image data of each whole slide is acquired using an image sensor having at least approximately 100 Megapixel resolution.
16. The non-transitory, computer-readable medium as set forth in claim 11 wherein the processor includes a GPU that performs the extracting of features and the operating of the attention network.
17. The non-transitory, computer-readable medium as set forth in claim 11 wherein the image data of each whole slide is acquired using an image sensor having at least approximately 100 Megapixel resolution.