Artificial intelligence system for determining clinical values through medical imaging
AssigneeUltrasound AI, Inc.
InventorRobert S. Bunn
Life of this patent
Estimating the term…
Abstract
Systems and methods for establishing a patient's current or future clinical or lab values are provided. A neural network is trained on a dataset of medical images, such as ultrasound images, that are tagged with information concerning the lab values of people who were imaged to produce the medical images. The trained neural network can then be provided with medical images of a patient, and the neural network can then make a determination as to the patient's current or future clinical or lab values.
Claims30
A system configured to make a medical determination, the system comprising:
- an image storage configured to store non-invasive images of a body part of a patient,
- the non-invasive images previously acquired via a process that does not require cutting or puncturing the patient's skin;
- image analysis logic comprising a trained neural network in communication with the image storage and configured to provide a determination regarding a clinical or lab value of the patient, the determination based on the stored non-invasive images and additionally based on a clinical data set;
- wherein the neural network comprises a layer configured to extract a set of feature representations from each of the non-invasive images;
- wherein the neural network comprises a layer configured to produce a classification from a set of feature representations;
- a user interface configured to provide the determination to a user, wherein the determination is based on the classification;
- and
- a processor configured to execute at least the image analysis logic.
A method for making a medical determination, the method comprising:
- obtaining a non-invasive image of a body part of a patient, the non-invasive image acquired without puncturing or cutting the patient's skin;
- providing the non-invasive image to a neural network that has been trained to determine current or future clinical or lab quantitative values from the image of the body part;
- passing the non-invasive image through one or more layers to produce a feature representation for each image;
- passing feature representations through one or more layers to produce a classification;
- and
- receiving the determination of the clinical or lab value based on the classification.
A method for making a medical determination, the method comprising:
- obtaining a non-invasive image of a body part of a patient, the non-invasive image acquired without puncturing or cutting the patient's skin;
- providing the non-invasive image to a neural network that has been trained to determine current or future clinical or lab quantitative values of the patient from the image of the body part;
- receiving the determination of the clinical or lab value;
- wherein the neural network has been trained using label smoothing, weight decay, a dropout layer, or another regularization technique;
- and
- wherein the neural network has been trained by decreasing a learning rate over time.
A method for training a neural network to make a medical determination, the method comprising:
- receiving a set of non-invasive images of body parts of a plurality of patients, the non-invasive images tagged with information concerning current or future lab values of the patient, the non-invasive images having been previously generated without puncturing or cutting the patient's skin;
- dividing the set of non-invasive images into a training set and a validation set;
- providing the images of the training set, and their tags, to a neural network to train the neural network to determine current or future, lab or clinical, quantitative values based on the non-invasive images of the body parts;
- providing images of a single patient from the validation set to the neural network to make the medical determination, and comparing the medical determination to the tag associated with the non-invasive images;
- and
- monitoring a validation accuracy on the validation set for a decrease in validation accuracy.
Dependencies are inferred from the claim text.