Large-scale physiological waveform retrieval via locality-sensitive hashing

Area of Science:

• Biomedical Informatics
• Signal Processing
• Machine Learning

Background:

• Retrieving similar physiological waveforms is crucial for clinical decision-making.
• Existing methods like k-nearest neighbor search are computationally intensive and slow for large datasets.
• Efficient similarity search in physiological time series data remains a challenge.

Purpose of the Study:

• To propose a fast and scalable locality-sensitive hashing (LSH) method for physiological waveform time series retrieval.
• To evaluate the performance of the proposed LSH method against naive k-nearest neighbor search.
• To demonstrate the method's effectiveness on real-world physiological data.

Main Methods:

• Implemented a locality-sensitive hashing (LSH) algorithm tailored for time series data.
• Compared the LSH method's retrieval time and accuracy against k-nearest neighbor (k-NN) search.
• Utilized an arterial blood pressure time series dataset from the MIMIC-II database for evaluation.

Main Results:

• The LSH method achieved up to an order of magnitude speed-up in retrieval time compared to k-NN.
• Retrieval accuracy remained high, at 95% or better, with the LSH method.
• The computational overhead for building the data structure was offset by query volume, with minimal additional memory cost.

Conclusions:

• The proposed LSH method offers a significant improvement in speed and scalability for physiological waveform similarity search.
• This approach provides a practical solution for analyzing large physiological time series datasets, such as those found in intensive care units.
• The method demonstrates high accuracy and efficiency, making it suitable for real-time applications in biomedical informatics.