Sepsis Prediction Using Temporal-Spatial Attention in ICU Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for sepsis prediction in ICU settings are inadequate for capturing complex temporal and spatial correlations in physiological data, leading to suboptimal diagnosis and prediction accuracy.
Innovation Solution
A computer-implemented method using a recurrent neural network (RNN) with gated recurrent units (GRU) to process physiological data, generating multi-dimensional feature encodings and computing short-range and spatial correlations through attention-based mechanisms to predict sepsis, involving data binning, self-attention, and key-query-value schemes for improved temporal and spatial correlations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods are used for sepsis prediction, then the system is simpler to implement, but the prediction accuracy is insufficient due to inability to capture complex temporal and spatial correlations
Solution Approach 1:
The patent segments the sepsis prediction task into multiple components: spatial correlation computation (capturing relationships between different physiological variables at the same time step), temporal correlation computation (capturing relationships between variables across different time steps), and feature encoding stages. This segmentation allows the complex prediction problem to be solved through modular, manageable components that can be implemented systematically.
Solution Approach 2:
The patent transforms the input physiological data into multi-dimensional feature encodings, adding dimensional depth to the data representation. By computing correlations across both spatial dimensions (different variables) and temporal dimensions (different time steps), the model captures complex patterns that cannot be detected in single-dimensional analysis, thereby improving prediction accuracy.
2Measurement precision
If complex correlation computations are performed, then prediction accuracy improves, but computational time and resources increase
Solution Approach 1:
The patent performs preliminary computations by pre-processing physiological data into multi-dimensional feature encodings before the actual prediction task. The spatial and temporal correlations are computed systematically as intermediate steps, preparing the data in advance for the final sepsis prediction. This preliminary action organizes complex computations into manageable stages, improving both accuracy and computational efficiency.
3Reliability
If more physiological variables are monitored, then prediction reliability improves, but data processing complexity increases
Solution Approach 1:
The patent employs a universal computational framework that can handle multiple physiological variables simultaneously. The same spatial correlation computation mechanism processes relationships between different variables (e.g., heart rate, blood pressure, temperature), and the same temporal correlation mechanism captures their evolution over time. This multi-functional approach allows reliable prediction across diverse variables without proportionally increasing processing complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method comprising: performing a prediction process comprising: based on input data comprising values of physiological measurement variables of a patient over a time period, computing first correlations in the input data, the computing comprising computing short range temporal correlations between values of physiological measurement variables at consecutive time steps using an attention-based mechanism and computing spatial correlations between values of different physiological measurement variables at a same time step using a self-attention mechanism; generating first updated node embeddings based on the input data and the first correlations, each node corresponding to a physiological measurement variable at a time step; using a recurrent neural network, RNN, updating the first updated node embeddings based on second correlations between the first updated node embeddings to generate temporally updated embeddings; and based on the temporally updated embeddings and using a neural network, NN, generating a prediction whether the patient will have sepsis.