Verily Life Sciences LLC

Ali Behrooz

Cheng-Hsun Wu

Charles Santori

Systems and methods for using a prediction model jointly with a normalization model to provide prediction results are provided. One example method includes receiving an input image of a tissue sample of a patient and generating a normalized image by applying a normalization model on the input image. The normalization model is configured to generate normalized data using input data for a prediction model, and the prediction model is configured to generate prediction results using normalized data generated by the normalization model. The normalization model and the prediction model are jointly trained. The method further includes generating a prediction of disease severity for the patient by applying the prediction model on the normalized image.

Auto-normalization for machine learning

One disclosed example method for simulated medical images using GANs includes receiving, by a generative adversarial network ('GAN'), a plurality of training images, the training images associated with an affliction and depicting different stages of progression for the affliction; generating, using the GAN, a latent space based on the training images, the latent space comprising a first set of data points indicating parameters associated with the training images; generating, using the GAN, a second set of data points in the latent space based on simulated images generated from the latent space, the simulated images based at least in part on the first set of data points; after generating the second set of data points: receiving a request for a first simulated image associated with the affliction; and generating and outputting, by the GAN, the first simulated image based on the latent space.

Systems and methods for simulating medical images using generative adversarial networks

Systems and methods for predicting images with enhanced spatial resolution using a neural network are provided herein. According to an aspect of the invention, a method includes accessing an input image of a biological sample, wherein the input image includes a first spatial resolution and a plurality of spectral images, and wherein each spectral image of the plurality of spectral images includes data from a different wavelength band at a different spectral channel; applying a trained artificial neural network to the input image; generating an output image at a second spatial resolution, wherein the second spatial resolution is higher than the first spatial resolution, and wherein the output image includes a fewer number of spectral channels than the plurality of spectral images included in the input image; and outputting the output image.

Transforming multispectral images to enhanced resolution images enabled by machine learning

Systems and methods of improving alignment in dense prediction neural networks are disclosed. A method includes identifying, at a computing system, an input data set and a label data set with one or more first parts of the input data set corresponding to a label. The computing system processes the input data set using a neural network to generate a predicted label data set that identifies one or more second parts of the input data set predicted to correspond to the label. The computing system determines an alignment result using the predicted label data set and the label data set and a transformation of the one or more first parts, including a shift, rotation, scaling, and/or deformation, based on the alignment result. The computing system computes a loss score using the transformation, label data and the predicted label data set and updates the neural network based on the loss score.

Shift invariant loss for deep learning based image segmentation

Various techniques are provided for performing automated full-cell segmentation and labeling in immunofluorescent microscopy. These techniques perform membrane segmentation and nuclear seed detection separate and independently from each other, then combine their results to identify cell boundaries. Some embodiments use texture- and kernel-based image processing to perform the method. In some embodiments, the method for obtaining membrane features disclosed herein can be used in conjunction with or separate from the nuclear features. The results can be used for a variety of purposes, including whole-area cell segmentation in fluorescence-based tissue imaging.

Growing community of inventors

San Mateo, CA, United States of America

Ali Behrooz