Three-dimensional contoured scanning photoacoustic imaging and virtual staining

A method for generating virtual histology images, the method comprising;

• using a first generator of a first generative adversarial network to generate a virtually stained histology image having a characteristic of hematoxylin and eosin staining, the first generator trained to generate the virtually stained histology image based on an input of a photoacoustic histology image;
• wherein the first generator is trained by:
• inputting a training photoacoustic histology image to the first generator, wherein the training photoacoustic histology image comprises a real image or a fake image;
• predicting an output image having the characteristic of hematoxylin and eosin staining;
• obtaining an initial metric value ranging from 1 to 0, the initial metric value representative of a likelihood that the predicted output image is real or fake generated by a first adversarial discriminator corresponding to the first generator;
• and
• updating the first generator such that a subsequent metric value generated by the first adversarial discriminator is closer to 0.5 than the initial metric value.

A method for generating virtual histology images, the method comprising;

• using a first generator of a first generative adversarial network to generate a virtually stained histology image having a characteristic of hematoxylin and eosin staining, the first generator trained to predict the virtually stained histology image based on an input of a photoacoustic histology image;
• wherein:
• the first generator comprises one or more first layers of the first generative adversarial network, the one or more first layers including one or more first convolutional layers, one or more first downsampling layers, one or more first upsampling layers, or a combination thereof;
• a first adversarial discriminator corresponding to the first generator comprises one or more second layers of the first generative adversarial network, the one or more second layers including one or more convolutional layers, one or more downsampling layers, or a combination thereof;
• and
• the first adversarial discriminator is trained to output a determination that the predicted virtually stained histology image is real or fake based on the one or more second layers.

A method for virtually staining a photoacoustic histology image, the method comprising:

• generating a virtually stained histology image using a first generative adversarial network trained to predict, from the photoacoustic histology image, the virtually stained histology image having a characteristic of hematoxylin and eosin staining;
• wherein:
• the first generative adversarial network comprises a first generator and a corresponding first adversarial discriminator;
• using the first generative adversarial network to predict the virtually stained histology image comprises using the first generator to predict the virtually stained histology image;
• the first adversarial discriminator comprises a classifier trained to output a determination that the predicted virtually stained histology image is real or fake;
• the first generator comprises one or more first layers of the first generative adversarial network, the one or more first layers including one or more first convolutional layers, one or more first downsampling layers, one or more first upsampling layers, or a combination thereof;
• the first adversarial discriminator comprises one or more second layers of the first generative adversarial network, the one or more second layers including one or more convolutional layers, one or more downsampling layers, or a combination thereof;
• and
• the output of the determination by the first adversarial discriminator is based on the one or more second layers.