Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12260946-B2
Publication Date
2025-03-25
Expiration Date
Abstract
An exemplary discovery platform includes machine-learning techniques for using medical imaging data to study a phenotype of interest, such as complex diseases with weak or unknown genetic drivers. An exemplary method of identifying a patient subgroup of interest, comprises inputting a plurality of medical images obtained from a group of clinical subjects into a trained unsupervised machine-learning model to obtain a plurality of embeddings in a latent space, clustering the plurality of embeddings to generate one or more clusters of embeddings, identifying one or more patient subgroups corresponding to the one or more clusters of embeddings, and associating each patient subgroup of the one or more patient subgroups with a covariant to identify the patient subgroup of interest.
Core Innovation
The invention provides a discovery platform for analyzing medical images from a group of clinical subjects. A plurality of medical images is input into a trained unsupervised machine-learning model to obtain a plurality of embeddings in a latent space, and the embeddings are clustered to generate one or more clusters of embeddings corresponding to one or more patient subgroups.
Each patient subgroup is associated with a covariant to identify the patient subgroup of interest. The covariant is selected to characterize and relate the discovered patient subgroups to phenotype-relevant factors described in the document, including treatment, disease progression, placebo vs treatment baselines and follow-ups, disease-progression values, and adverse side effects.
The document also describes downstream analyses tied to the platform outputs, including association analyses linking patient subgroups to covariates and phenotypes, and determining a correlation metric, including a P value, compared to a predefined threshold for significant association. It further describes predicting continuous medical diagnosis scores from embeddings via linear or linear-mixed models, relating genetic variants to disease using P value thresholds, and evaluating treatment effects by imputing drug-response phenotypes from progression embeddings and generating simulated images via conditional GAN for interpretability.
Claims Coverage
The independent claims in the provided set are directed to identifying a patient subgroup of interest from medical images by generating latent-space embeddings using a trained unsupervised machine-learning model, clustering the embeddings, and associating the resulting patient subgroups with a covariant. The independent claims share four main inventive features.
Latent-space embedding generation from medical images using a trained unsupervised machine-learning model
Inputting a plurality of medical images obtained from a group of clinical subjects into a trained unsupervised machine-learning model to obtain a plurality of embeddings in a latent space.
Clustering latent-space embeddings into embedding clusters
Clustering the plurality of embeddings to generate one or more clusters of embeddings.
Patient subgroup identification corresponding to embedding clusters
Identifying one or more patient subgroups corresponding to the one or more clusters of embeddings.
Covariant-based association to identify a patient subgroup of interest
Associating each patient subgroup of the one or more patient subgroups with a covariant to identify the patient subgroup of interest.
Across the independent claims, the inventive scope centers on unsupervised latent embedding of medical images, clustering those embeddings, mapping clusters to patient subgroups, and using a covariant association to determine which patient subgroup is of interest.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Applied to complex diseases such as NASH using biopsy (H&E) data across clinical trials.
Interested in licensing this patent?