Automated assessment of wound tissue
AssigneeAstraZeneca AB
InventorsYinhai WANG, Adrian Mark FREEMAN
Life of this patent
Estimating the term…
Abstract
A method of assessing a wound in a subject is provided. The method comprises obtaining one or more optical coherence tomography images of the wound and analysing the one or more optical coherence tomography images using a deep learning model that has been trained to classify pixels in an optical coherence tomography image of a wound between a plurality of classes comprising a plurality of classes associated with different types of wound tissue, thereby obtaining for each image analysed, an indication of the location of tissue likely to belong to each of the different types of wound tissue in the respective image.
Claims15
A method of assessing a wound in a subject, the method comprising:
- obtaining one or more optical coherence tomography images of the wound;
- and
- analysing the one or more optical coherence tomography images using a deep learning model that has been trained to classify pixels in an optical coherence tomography image of a wound between a plurality of classes comprising a plurality of classes associated with different types of wound tissue, thereby obtaining for each image analysed, an indication of the location of tissue likely to belong to each of the different types of wound tissue in the respective image.
The method of any preceding claim, wherein:
- the deep learning model has been trained using a plurality of training optical coherence tomography images, wherein areas of each training image showing visual features indicative of the presence of the different types of wound tissues are labelled accordingly;
- and/or
- the deep learning model takes as input a single image and analysing the one or more optical coherence tomography images comprises providing each of the one or more optical coherence tomography images individually as input to the deep learning model.
The method of any preceding claim, wherein the indication of tissue likely to belong to each of the different types of wound tissue in the respective image is obtained as one or more segmentation maps, wherein a segmentation map is an image of the same size as the image analysed, with pixels classified in a particular class assigned a different value from pixels that have not been classified in the particular class.
The method of any preceding claim, wherein:
- each optical coherence tomography image of the wound shows signal from the surface of the skin of the subject to a maximum depth, optionally wherein the maximum depth is between 1 and 2 mm;
- and/or
- wherein a plurality of optical coherence tomography images of the wound are obtained and analysed, together forming a stack of images that show signal across an area of the surface of the skin of the subject, optionally wherein the method further comprises combining the indications of the location of the tissue likely to belong to each of the different types of wound tissue, in the respective images to obtain a three-dimensional map of the location of tissue likely to belong to each of the different types of wound tissue.
The method of any preceding claim, wherein the deep learning model is a convolutional neural network, and/or wherein the deep learning network is a u-net or a generative adversarial network, and/or wherein the deep learning network comprises a contracting path that reduces spatial information and increases feature information, and an expansive path that combines features and spatial information, optionally wherein the contracting path comprises convolution layers followed by ReLU and max pooling layers, and the expansive path comprises a sequence of up-convolutions and concatenations with features from the contracting path.
The method of any preceding claim, further comprising applying one or more post-processing steps to the output of the deep learning model, optionally wherein the post-processing steps comprise one or more of:
- smoothing the boundaries of the areas comprising pixels identified as belonging to one or more classes, and re-labelling pixels identified as belonging to one or more classes where the pixels satisfy one or more criteria applying to the neighbouring pixels.
The method of any preceding claim, further comprising determining, using the output from the deep learning model, the surface area corresponding to the pixels identified by the deep learning model as likely to belong to at least one of the different types of wound tissue in the respective image, optionally comprising determining one or more of:
- the surface area corresponding to the pixels identified by the deep learning model as likely neoepidermis, the surface area corresponding to the pixels identified by the deep learning model as likely clot tissue, the surface area corresponding to the pixels identified by the deep learning model as likely granulation tissue, in at least one of the one or more images.
The method of any preceding claim, further comprising:
- (i) determining the volume of at least one of the different types of wound tissue in the wound, by:
- analysing a plurality of images of optical coherence tomography images of the wound using the deep learning model;
- determining, using the output form the deep learning model, for each of the plurality of images, the surface area corresponding to the pixels identified as likely to belong to the respective one of the different types of wound tissue, such as the surface area corresponding to the pixels identified as likely neoepidermis, the surface area corresponding to the pixels identified by the deep learning model as likely clot tissue, and/or the surface area corresponding to the pixels identified by the deep learning model as likely granulation tissue;
- and
- multiplying the determined surface area(s) in each image by a predetermined distance;
- and/or
- (ii) determining the width of the wound based on a dimension of the location(s) of tissue identified as likely to belong to one or more of the different types of wound tissue in at least one of the one or more images, optionally wherein the one or more of the different types of wound tissue include neoepidermis, clot and granular tissue.
The method of any preceding claim, wherein the subject is a human subject and/or wherein the wound is a skin wound, and/or wherein the wound is a traumatic wound, a surgical wound, or a skin ulcer.
Dependencies are inferred from the claim text.