Infusion Pump Misload Detection via Machine Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional misload detection methods for infusion pumps, such as electromechanical switches, are unreliable and prone to false positives/negatives due to the flexibility of infusion set placement, and introduce complexity and contamination risks.

Innovation Solution

A machine learning-based system that uses images from cameras to identify nonconformities in the infusion set placement by comparing against correctly loaded set images, preventing infusion and generating corrective messages when misloads are detected.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If electromechanical switches are used for misload detection, then the pump can detect infusion set placement errors, but the system becomes more complex and prone to false positives/negatives due to the flexibility of infusion set placement

Engineering Contradiction:
Improvemisload detection reliabilityVSAvoidpump system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces electromechanical switches with a machine learning-based image recognition system. Cameras capture images of the infusion set placement, and a trained machine learning model analyzes these images to detect misloads. This substitution eliminates the mechanical complexity of switches and their wiring while improving reliability by using flexible image-based detection that can handle various infusion set placements.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system uses visual copying of the infusion set placement through camera images instead of direct mechanical sensing. The machine learning model learns from training images of correct and incorrect placements, creating a digital representation of proper assembly that can be compared against new images to detect errors without physical contact.

Inventive Principle:
Principle #26Copying

2Reliability

If electromechanical switches are used for misload detection, then the pump can identify placement errors, but the risk of contamination increases due to additional components

Engineering Contradiction:
Improvemisload detection accuracyVSAvoidcontamination risk
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

By replacing electromechanical switches with optical cameras and machine learning analysis, the patent eliminates additional mechanical components that could introduce contamination. The image-based detection method is non-contact and does not require physical sensors within the fluid path or assembly area, reducing contamination risks while maintaining detection accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If machine learning-based image recognition is used for misload detection, then false alarms are reduced and detection reliability is enhanced, but the computational requirements and processing time increase

Engineering Contradiction:
Improvemisload detection reliabilityVSAvoidimage processing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The machine learning model is trained in advance on extensive datasets of correct and incorrect infusion set placements. This preliminary training creates a pre-configured recognition system that can quickly classify new images without requiring complex real-time computation. The heavy computational work is done beforehand during the training phase, not during actual pump operation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4156195B1Machine learning enabled detection of infusion pump misloads
Publication Date: 2024.10.30 CAREFUSION 303 INC
  • EP4156195B1 patent drawingFigure 1A
  • EP4156195B1 patent drawingFigure 1B
  • EP4156195B1 patent drawingFigure 2A

AI summary

A method may include capturing, by a camera at an infusion pump, one or more images of the pump loaded with an infusion set. A machine learning model may be applied to the images to detect nonconformities that may be present in the images of the pump loaded with the infusion set. Examples of nonconformities include a misload of the intravenous set in which an upper fitment, a lower fitment, and/or a tubing of the intravenous set is misplaced within the pump. In response to an output of the machine learning model indicating a presence of a nonconformity in the one or more images of the pump loaded with the infusion set, a corrective action may be performed. For example, the pump may be prevented from performing an infusion and a message identifying the nonconformities may be generated. Related methods and articles of manufacture are also disclosed.