Machine learning to predict lung cancer involvement of lymph nodes
AssigneeJohnson & Johnson Enterprise Innovation Inc.
InventorsGeorge R. Washko, Jr., Raul San Jose Estepar, Charles Matthew Kinsey, Christopher Scott Stevenson
Life of this patent
Estimating the term…
Abstract
Disclosed herein are methods for determining a subject level risk of metastatic cancer involving the training and/or deployment of models to determine 1) a lymph node level risk of individual lymph node involvement and/or 2) a subject level risk of lymph node involvement. Thus, the methods can identify patients who are high or low risk for having nodal disease and optionally enable the guided intervention of cancer patients, for example, via treatment.
Claims20
A method of determining a subject level risk of metastatic cancer of a subject, the method comprising:
- obtaining one or more images captured from the subject comprising a plurality of lymph nodes of the subject;
- and
- predicting the subject level risk of metastatic cancer by applying a risk model to extracted features of the obtained one or more images, the risk model trained to predict lymph node (LN) level risks of metastatic cancer for lymph nodes in images,
- wherein the risk model is trained using at least labels derived from imputed LN-level risks of metastatic cancer for a set of lymph nodes of a reference individual, the imputed LN-level risks of the set of lymph nodes determined using at least an imputation model that discriminates between cancerous and non-cancerous lymph nodes of training images.
A method of training a risk model, the method comprising:
- applying an imputation model to generate imputed LN-level risks for a plurality of lymph nodes in training images;
- selecting a set of lymph nodes, wherein the lymph nodes in the set have LN-level risks that are greater than a median risk value;
- and
- using the LN-level risks of the lymph nodes in the set of lymph nodes as reference ground truths for training the risk model, the risk model predicting LN-level risks with a greater range of probabilities than a range of probabilities of imputed LN-level risks predicted by the imputation model.
A system for determining a subject level risk of metastatic cancer of a subject, the system comprising:
- an imaging device configured to capture one or more images of the subject;
- and
- a computing device configured to perform the steps of:
- obtaining one or more images captured from the subject comprising a plurality of lymph nodes of the subject;
- and
- predicting the subject level risk of metastatic cancer by applying a risk model to extracted features of the obtained one or more images, the risk model trained to predict lymph node (LN) level risks of metastatic cancer for lymph nodes in images,
- wherein the risk model is trained using at least labels derived from imputed LN-level risks of metastatic cancer for a set of lymph nodes of a reference individual, the imputed LN-level risks of the set of lymph nodes determined using at least an imputation model that discriminates between cancerous and non-cancerous lymph nodes of training images.
Dependencies are inferred from the claim text.