Systems and methods for training a model to predict survival time for a patient
Inventors
Beck, Andrew H. • Khosla, Aditya
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11017532-B1
Publication Date
2021-05-25
Expiration Date
Abstract
In some aspects, the described systems and methods provide for a method for training a model to predict survival time for a patient. The method includes accessing annotated pathology images associated with a first group of patients in a clinical trial. Each of the annotated pathology images is associated with survival data for a respective patient. Each of the annotated pathology images includes an annotation describing a tissue characteristic category for a portion of the image. Values for one or more features are extracted from each of the annotated pathology images. A model is trained based on the survival data and the extracted values for the features. The trained model is stored on a storage device.
Core Innovation
The invention trains one or more models to predict survival time for a patient using a plurality of annotated pathology images associated with a group of patients in a randomized controlled clinical trial. Each annotated pathology image is associated with survival data for a respective patient and includes at least one annotation describing a tissue characteristic category for a portion of the image. The annotated pathology images include a first plurality associated with a first group of patients in a first treatment group and a second plurality associated with a second group of patients in a second treatment group.
The invention trains the one or more models based on the survival data and the plurality of annotated pathology images such that the one or more trained models include a first plurality of features based in part on the first plurality of annotated pathology images and a second plurality of features based in part on the second plurality of annotated pathology images. The trained models are stored on at least one storage device.
The invention enables separate prognostic performance and comparison of prognostic performance between the first and second treatment groups using predicted survival data from each group. Prognostic specificity reflects likelihood of correctly identifying a subset of patients that respond to a particular treatment, and subset survival analyses are performed using respective predicted survival data from the first and second treatment groups during the comparing step.
Claims Coverage
The document includes three independent claims, all centered on training survival-prediction models from annotated pathology images in randomized controlled clinical trials with treatment-group-specific feature sets. Dependent claims further define group-specific prognostic performance, prognostic specificity, subset survival analyses, histopathology-derived features, and example model classes.
Training survival-time models from treatment-group annotated pathology images
Accessing a plurality of annotated pathology images associated with a group of patients in a randomized controlled clinical trial, where each annotated pathology image is associated with survival data and includes at least one annotation describing a tissue characteristic category for a portion of the image, and the plurality includes a first plurality associated with a first treatment group and a second plurality associated with a second treatment group.
Treatment-group feature set training and storing trained models
Training one or more models based on the survival data and the plurality of annotated pathology images, where the one or more trained models include a first plurality of features based in part on the first plurality of annotated pathology images and a second plurality of features based in part on the second plurality of annotated pathology images, and storing the one or more trained models on at least one storage device.
Specificity and subset survival analysis using predicted survival data
Determining first prognostic performance and second prognostic performance for the first and second treatment groups by comparing predicted survival data with respective survival data, determining specificity of the prognostic power by comparing the first prognostic performance and the second prognostic performance, and performing subset survival analyses using respective predicted survival data from the first and second treatment groups as part of the comparing step.
Histopathology-derived feature specification and trained model classes
Selecting the first plurality of features and/or the second plurality of features from specified histological and spatial metrics such as areas of tissue components, immune-cell areas, mitotic figures, nuclear grade, and distances between cellular components, and where the one or more trained models include a generalized linear model, a random forest, a support vector machine, and/or a gradient boosted tree.
Downstream prediction using treatment-group feature sets
Using the one or more trained models to process at least the first plurality of features to predict survival data for patients in the first group and to process at least the second plurality of features to predict survival data for patients in the second group.
The claims define training survival-prediction models from randomized controlled clinical trial annotated pathology images where tissue characteristic category annotations support treatment-group-specific feature sets, with the trained models stored. Dependent claim coverage further specifies separate prognostic performance by treatment group, prognostic specificity, subset survival analyses, histopathology-derived feature metrics, and example model classes.
Stated Advantages
Enables prognostic specificity for identifying a subset of patients that respond to a particular treatment.
Allows separate and comparable prognostic performance assessment for first and second treatment groups using predicted survival data.
Supports subset survival analyses using predicted survival data from each treatment group.
Documented Applications
Selecting candidate patients for trials using the trained survival-prediction approach described for randomized controlled clinical trial settings.
Interested in licensing this patent?