Machine Learning Model for Subject Position Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Immobile or partially immobile subjects on person support apparatuses are at risk of pressure injuries and other conditions if left in an immobile state for extended periods, and certain subject positions may indicate emergencies that require immediate clinician attention, necessitating remote monitoring of subject positioning.

Innovation Solution

A system utilizing cameras and machine learning models trained with labeled subject and synthetic data to predict subject positions on a person support apparatus, incorporating data from various sensors like infrared images and load sensors, to remotely monitor and alert for potentially dangerous positions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual monitoring of subject positioning is used, then system complexity is low, but monitoring precision and reliability are insufficient

Engineering Contradiction:

Improvesubject position detection accuracyVSAvoidmonitoring system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces manual visual inspection with automated image processing and machine learning algorithms. Cameras capture images of subjects on the support apparatus, and a trained machine learning model automatically analyzes these images to detect subject positions and potential pressure injury risks, eliminating the need for continuous manual monitoring while significantly improving detection accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent uses synthetic training data - computer-generated images that replicate real subject positions and conditions - to train the machine learning model. This synthetic copying of real-world scenarios enables the model to learn accurate position detection without requiring extensive manual annotation of actual medical images, reducing complexity while maintaining high precision.

Inventive Principle:

Principle #26Copying

2Reliability

If continuous monitoring is implemented, then subject safety is improved, but energy consumption increases

Engineering Contradiction:

Improvesubject safety monitoringVSAvoidsystem energy consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system implements periodic monitoring by capturing images at specified intervals rather than continuously streaming video. The machine learning model processes these periodic images to detect subject positions and changes, maintaining safety monitoring reliability while significantly reducing energy consumption compared to continuous real-time processing.

Inventive Principle:

Principle #19Periodic action

Solution Approach 2:

The machine learning model performs self-service by automatically analyzing captured images without requiring continuous human intervention. The system autonomously detects subject positions, identifies potential risks, and can trigger alerts only when necessary conditions are met, reducing energy consumption while maintaining reliable safety monitoring.

Inventive Principle:

Principle #25Self-service

3Measurement precision

If extensive training data is used, then model accuracy is improved, but data processing time increases

Engineering Contradiction:

Improveposition prediction accuracyVSAvoidmodel training time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent generates synthetic training data by copying and transforming real image characteristics into artificial images with known ground truth labels. This synthetic data multiplication provides extensive training examples without requiring proportional increases in manual annotation time, as the synthetic generation process is automated and rapid compared to manual labeling.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary data preparation by generating and organizing synthetic training data before model training begins. This preliminary action includes creating diverse synthetic images representing various subject positions and conditions, which are then ready for immediate model training, reducing the overall time required during the actual training phase.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20240394907A1Devices, systems, and methods to remotely monitor subject positioning

Publication Date: 2024.11.28 HILL ROM SERVICES INC

AI summary

A method includes receiving subject training data comprising a plurality of images of subjects in a plurality of positions on a person support apparatus, labeling the plurality of images based on the positions of the subjects to generate labeled subject training data, generating synthetic training data comprising computer generated images of artificial subjects in a plurality of positions on an artificial person support apparatus, labeling the synthetic training data based on the positions of the artificial subjects to generate labeled synthetic training data, and training a machine learning model based on the labeled subject training data and the labeled synthetic training data, using supervised learning techniques, to generate a trained model to predict a subject position based on an image of the subject.