Alarm Annotation Workflow for Actionable Alert Discrimination

Solution Overview

Problem

Medical workers face a significant burden due to the high frequency of non-actionable alarms from medical devices, leading to 'alarm fatigue', and existing systems lack the capability to accurately discriminate between actionable and non-actionable alerts without requiring extensive training data.

Innovation Solution

A system and method that allows medical workers to input annotations through an interface to classify alerts as actionable or non-actionable, facilitating the collection of high-quality training data for an inference engine, thereby reducing the burden and improving alert discrimination.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If an inference engine with machine learning is used to discriminate actionable alarms from non-actionable alarms, then the discrimination accuracy is improved, but an enormous amount of training data is required

Engineering Contradiction:

Improvediscrimination accuracyVSAvoidtraining data volume

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The system performs preliminary action by having medical workers manually annotate alarms during routine operations. These annotations are collected and stored as training data before being used to train the inference engine. This preliminary data collection phase enables the system to build up sufficient training data over time without requiring a large initial dataset, thereby resolving the contradiction between needing high discrimination accuracy and requiring enormous training data volume.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If medical workers manually check each alert to determine if it is actionable, then the discrimination accuracy is improved, but the workload and time consumption increase significantly

Engineering Contradiction:

Improvediscrimination accuracyVSAvoidtime for alert management

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary action by having medical workers manually annotate alarms during routine operations. These annotations are collected and stored as training data before being used to train the inference engine. This preliminary data collection phase enables the system to build up sufficient training data over time without requiring a large initial dataset, thereby resolving the contradiction between needing high discrimination accuracy and requiring enormous training data volume.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables self-service by automatically training the inference engine using collected annotations and then autonomously discriminating between actionable and non-actionable alarms. Once the system is set up, it serves itself by continuously improving its discrimination capability without requiring ongoing manual review of each alarm, thereby reducing time loss while maintaining high accuracy.

Inventive Principle:

Principle #25Self-service

3Reliability

If the number of sensors and medical devices for monitoring is increased, then the monitoring capability is improved, but the number of non-actionable alarms and alarm fatigue increase

Engineering Contradiction:

Improvemonitoring capabilityVSAvoidalarm fatigue

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The system implements feedback by using the inference engine to analyze alarm patterns and provide information about whether alarms are actionable or non-actionable. This feedback loop enables the system to learn from past alarms and improve its discrimination capability over time, thereby reducing alarm fatigue while maintaining reliable monitoring capability through increased sensor deployment.

Inventive Principle:

Principle #23Feedback

Data Source

PatentEP4675637A1Annotation collecting system, and annotation collecting method

Publication Date: 2026.01.07 NIHON KOHDEN CORP

Figure 1
Figure 2
Figure 3~4

AI summary

A medical device is configured to issue an alert. An annotation assigning device is configured to provide an input user interface arranged to input managing information indicating whether the alert is an actionable alarm or a non-actionable alarm. The annotation assigning device is configured to assign an annotation corresponding to the managing information to the alert.