Artificial Neural Network Bio Event Prediction With Dual Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting bio events using deep learning models provide limited information, making it difficult to determine the criticality of potential events and prioritize necessary interventions for patients at risk.
Innovation Solution
An artificial neural network model is employed to generate a first output indicating the expected time of occurrence and a second output indicating the possibility of a bio event, enabling comprehensive prediction information that aids in determining patient priority based on bio information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional deep learning models are used to predict bio events, then the prediction can be made, but only very limited information is provided making it difficult to determine criticality and prioritize interventions
Solution Approach 1:
The patent segments the prediction task into two distinct output components: (1) a binary classification output indicating whether a bio event will occur within a predetermined time period, and (2) a regression output predicting the expected time of occurrence. This segmentation allows the model to provide comprehensive prediction information (completing the information loss) while maintaining a relatively simple neural network structure (avoiding excessive complexity).
Solution Approach 2:
The patent adds a temporal dimension to the traditional binary classification by introducing a continuous time prediction output. Instead of only predicting whether an event occurs (0/1), the model now predicts both the occurrence probability and the expected time of occurrence, transforming the output space from one-dimensional to two-dimensional and thereby providing more comprehensive prediction information.
2Reliability
If multiple patients with potential bio events are monitored, then all patients can be identified, but it is difficult to know which one is more critical and to take necessary measures depending on critical condition
Solution Approach 1:
The patent performs preliminary action by having the neural network model automatically generate comprehensive prediction information (expected time of occurrence and occurrence probability) for all patients simultaneously. This pre-computed information includes the criticality assessment, so when multiple patients need monitoring, the system already has the prioritization data ready, eliminating the need for time-consuming manual assessment and enabling immediate identification of critical cases.
Solution Approach 2:
The patent introduces the expected time of occurrence as an intermediary metric that bridges the gap between raw bio information and criticality assessment. This intermediary output provides a quantitative measure of urgency, allowing clinicians to objectively prioritize patients based on predicted time of event occurrence rather than subjective judgment, thereby improving reliability of priority determination and reducing assessment time.
Data Source
Figure 1~2
Figure 3
Figure 4A~4C
AI summary
According to an embodiment of the present disclosure, disclosed is a method for predicting a bio event. Specifically, according to the present disclosure, a computing device obtains bio information of a patient; and generates a first output indicating an expected time of occurrence of the bio event based on the bio information by using a pre-trained artificial neural network model.