Osteoporosis Risk Prediction: A Novel AI-Powered Tool

This invention is a machine learning-based computational model and clinical decision support tool designed to assess bone health using pelvic X-rays. Its most significant advantage is providing a binary bone mineral density (BMD) estimate to patients who have radiograph images available, offering an additional opportunity for risk assessment and management.

Description

The technology functions by taking a pelvic X-ray as input and outputing a simulated binary dual-energy X-ray absorptiometry (DEXA) score. This score classifies the patient's bone mineral density as "normal" or "at risk of osteoporosis," emulating the gold standard in clinical care. The model aims to provide valuable BMD estimates to radiologists, clinicians, and patients, especially when DEXA scans are not immediately feasible, such as in acute or subacute settings. This approach addresses the current subjectivity and variability in osteoporosis assessment by radiologists using radiographic images.

Applications

- Emergency departments and urgent care centers for initial patient evaluations.
- Outpatient clinics for follow-up and ongoing patient management.
- Integration into pelvic radiograph imaging interpretation workflows for radiologists.
- Primary care settings for general health screenings and risk assessments.
- Telemedicine platforms to extend bone health assessments to remote patients.

Advantages

- Provides osteoporosis risk assessment from pre-existing pelvic X-rays, eliminating the need for a separate DEXA scan appointment.
- Offers an objective, computational model to assess bone health, reducing subjective interpretation and variability among radiologists.
- Increases opportunities for early osteoporosis risk assessment and management, potentially decreasing patient morbidity and mortality.
- Utilizes readily available and less costly X-ray imaging compared to CT scans or dedicated DEXA appointments.
- There are currently no other inventions that extrapolate DEXA scores directly from pelvic radiographs.

Invention Readiness

A prototype model has been developed and is based on data from 582 patients who had both pelvic X-rays and DEXA scans performed within 180 days. The model currently uses 93 radiomics features from the segmented hip region on pelvic X-rays and has an area-under-the-curve (AUC) value of 0.65. Further refinement of the model and gathering more data are anticipated to achieve better results, with the ultimate vision of incorporating the validated software model into clinical practice within pelvic radiograph imaging interpretation workflows.

IP Status

Copyright