Time Series Deep Survival Analysis with Active Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional survival analysis models, including Cox proportional hazards and deep survival neural networks, face challenges in handling high-dimensional, sparse, and time-series medical data with a high censoring ratio, particularly in predicting survival times from electronic health records, due to limitations in utilizing time series features and dealing with incomplete data.
Innovation Solution
A time series deep survival analysis system combined with active learning, which includes a data collection module, an active learning module, and a time series deep survival analysis module, uses a novel sampling strategy to label right-censored data, constructs a neural network model with bidirectional LSTM and attention mechanisms, and employs a multi-category tag classification approach to predict survival times, effectively utilizing time series features and improving model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional Cox proportional hazards model or deep survival neural network is used, then the model structure is simple or can handle non-linear relationships, but the model cannot effectively utilize time series features and performs poorly with high censoring ratios
Solution Approach 1:
The patent combines multiple previously separate components into a unified model: (1) merging time series feature extraction capabilities with survival analysis, (2) combining active learning mechanisms with deep neural networks, and (3) integrating multiple loss functions to simultaneously handle censored and uncensored data. This unified approach enables the model to effectively process time series medical data with high censoring ratios while maintaining prediction accuracy
Solution Approach 2:
The patent introduces dynamic elements into the survival analysis model: (1) using bidirectional LSTM to dynamically capture temporal dependencies in both forward and backward directions, (2) implementing active learning that dynamically selects which censored samples to label based on model uncertainty, and (3) employing dynamic loss weighting that adapts during training to balance contributions from different data types
2Quantity of substance
If more right-censored data is used for training, then the model can utilize more data, but the model performance deteriorates due to incomplete survival information
Solution Approach 1:
The patent transforms the previously harmful effect of right-censored data (incomplete survival information that degraded model performance) into a beneficial training resource. By designing specialized loss functions and active learning strategies, the model now effectively utilizes censored samples to learn temporal patterns and feature representations, converting data incompleteness into a source of additional training signal that improves overall model performance
Solution Approach 2:
The patent changes the parameter representation and processing of censored data: (1) introducing interval-based survival time representation instead of point estimates, (2) modifying loss function parameters to account for censoring uncertainty, and (3) adjusting sampling probabilities in active learning based on censoring status. These parameter changes enable effective utilization of censored data without sacrificing accuracy
3Measurement precision
If active learning is applied to label right-censored data, then the labeled data quality improves, but the data processing time and complexity increase
Solution Approach 1:
The patent applies partial active learning by selectively labeling only a subset of right-censored data rather than attempting to label all censored samples. The active learning mechanism identifies and prioritizes the most informative censored samples for labeling, achieving significant performance improvements with minimal additional labeling effort and processing time
Solution Approach 2:
The patent implements a self-service active learning system where the model automatically identifies which censored samples would be most beneficial to label, ranks them by informational value, and guides the labeling process without requiring extensive manual intervention. This automation reduces the time and human resources needed for data processing while maintaining high data quality
Data Source
AI summary
Provided is a time series deep survival analysis system combined with active learning. The system includes: a data collection module, an active learning module, and a time series deep survival analysis module; the data collection module is used for obtaining survival data of objects to be analyzed; combined with an active learning method, the active learning module selects a part of right censored data to label a survival time; and the time series deep survival analysis module constructs a time series deep survival analysis neural network model, and takes uncensored data and right censored data as model inputs, so as to obtain survival time prediction results of the objects to be analyzed. The present application can make full use of the right censored data in the survival data and time series features.