Liver fat quantification from DEXA data

Inventors

Amar, David • Albright, Jack • Probert, Christopher • Mukherjee, Sumit • Koller, Daphne

Assignees

Insitro Inc

Abstract

An exemplary method for predicting one or more adipose depots for a patient includes receiving one or more Dual-energy X-ray Absorptiometry (DEXA) scans comprising at least a portion of a torso of the patient; providing at least one or more portions of the one or more DEXA scans to a trained machine-learning model, wherein the machine-learning model is trained using a training dataset comprising: a plurality of training DEXA scans of a plurality of subjects and a plurality of corresponding Magnetic Resonance Imaging (MRI)-image-based adiposity scores of the plurality of subjects; and predicting the one or more adipose depots for the patient utilizing the trained machine-learning model.

Core Innovation

The invention provides a system and corresponding medium and method for predicting one or more adipose depots for a patient using Dual-energy X-ray Absorptiometry (DEXA) scans. The system receives one or more DEXA scans comprising at least a portion of a torso of the patient and provides at least one or more portions of the DEXA scans to a trained machine-learning model at inference time.

The trained machine-learning model is trained using a training dataset that links training DEXA scans with Magnetic Resonance Imaging (MRI)-image-based adiposity scores. The training dataset comprises a plurality of training DEXA scans of a plurality of subjects and a plurality of MRI-image-based adiposity scores of the plurality of subjects, wherein the adiposity scores are based on a plurality of MRI images of the plurality of subjects.

The disclosed approach targets adipose depot prediction, including representation of liver fat using PDFF and optional MRI-derived adiposity score types. The training labels can include liver fat estimates and/or estimated visceral adipose tissue (VAT), estimated subcutaneous adipose tissue (SAT), and estimated gluteofemoral adipose tissue (GFAT) scores.

Claims Coverage

Independent claims cover the end-to-end DEXA-to-adipose-depot prediction pipeline using a trained machine-learning model whose training uses paired DEXA scans and MRI-image-based adiposity scores (3 inventive features).

Across the independent claims, the invention is consistently defined by receiving torso-portion DEXA scans, applying them to a trained machine-learning model trained with paired DEXA scans and MRI-image-based adiposity scores, and predicting one or more adipose depots for the patient.

Stated Advantages

Documented Applications