Determining scores indicative of times to events from biomedical images

Inventors

Muhammad, Hassan • Xie, Chensu

Assignees

Memorial Sloan Kettering Cancer Center

Abstract

Presented herein are systems and methods for determining scores from biomedical images. A computing system may identify a plurality of tiles in a first biomedical image derived from a sample of a subject. Each tile may correspond to features of the sample. The computing system may apply the plurality of tiles to a machine learning (ML) model. The ML model may include: an encoder to generate a plurality of feature vectors based on the plurality of tiles; a clusterer to select a subset from the plurality of feature vectors; and an aggregator to determine a first score indicative of a time to an event for the subject resulting from the features of the sample. The model may be trained in accordance with a loss derived from second scores determined for second biomedical images. The computing system may store an association between the score and the first biomedical image.

Core Innovation

The invention provides an end-to-end biomedical-imaging risk model that determines a first score from biomedical images derived from a subject. A plurality of tiles in a first biomedical image are identified, where each tile corresponds to one or more features of a sample from the subject, and the tiles are applied to a machine learning model including an encoder to generate a plurality of feature vectors.

Within the machine learning model, a clusterer defines a plurality of centroids in a feature space to select a subset of feature vectors from the plurality of feature vectors. An aggregator having a second plurality of weights combines the selected subset of feature vectors to determine a first score indicative of a time to an event for the subject resulting from the one or more features of the sample.

The model is trained in accordance with a loss derived from a second plurality of scores determined for a corresponding second plurality of biomedical images. Training updates at least one of the encoder weights, the clusterer centroids, or the aggregator weights, and learned parameters are stored. The system stores an association between the score and the first biomedical image to support risk assessment.

Claims Coverage

The partial claim set includes three independent claims: a determining method, a training method, and a system. Across these independent claims, there are six inventive features covering tile-to-feature-vector encoding, centroid-based subset selection, weighted aggregation to a time-to-event score, loss-based training using corresponding scores, storing a score-image association, and a computing system configured for the same pipeline.

Across the independent claims, the invention centers on an ML pipeline where tiles from biomedical images are encoded into feature vectors, selected via centroid-defined clustering in a feature space, and aggregated with learned weights to output a score indicative of time to an event. Training uses a loss derived from corresponding scores for other biomedical images, and the system stores an association between the produced score and the first biomedical image.

Stated Advantages

Documented Applications