Optical Flow Autoencoder for Movement Biomarker Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for tracking and predicting movement, especially small and irregular changes, are inefficient and subjective, making it difficult to capture and categorize movement-based biomarkers effectively, such as tremors, for disease diagnosis.
Innovation Solution
A method involving video analysis using optical flow extraction and autoencoders to generate movement-based biomarkers, where optical flows are processed to encode tremor frequencies and types, and an adversarial autoencoder network is trained to improve accuracy and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional computer vision or human raters are used to track and predict movement, then the system is simple to implement, but the measurement precision is insufficient to capture small and irregular changes in movement
Solution Approach 1:
The patent replaces traditional mechanical/optical motion tracking systems and human visual assessment with a computational approach using optical flow algorithms and neural networks. This substitution enables precise detection of small and irregular movements through pixel-level analysis of video frames, achieving high measurement precision while maintaining system simplicity through software-based implementation.
2Reliability
If traditional movement tracking methods are used, then the device complexity is low, but the reliability is insufficient to accurately categorize movement patterns for disease diagnosis
Solution Approach 1:
The patent introduces optical flow vectors as an intermediary representation between raw video frames and movement categorization. These vectors encode pixel displacement information that serves as a reliable bridge for analyzing movement patterns. The neural network then processes these intermediate representations to accurately categorize movement types, achieving high reliability in disease diagnosis while managing system complexity through this structured intermediate step.
Solution Approach 2:
The patent transforms 2D video frame data into optical flow fields that add a temporal dimension to movement analysis. By computing pixel displacements between consecutive frames, the system creates a third dimension of information (motion vectors) that enables more reliable categorization of movement patterns, including subtle tremors and irregular motions associated with neurological conditions.
3Measurement precision
If extensive training data is used to train the autoencoder, then the measurement precision improves, but the loss of time for data processing and training increases
Solution Approach 1:
The patent performs preliminary action by pre-training the autoencoder model on a large dataset of optical flow sequences to learn general movement patterns and biomarker representations. This pre-training phase captures essential features of normal and abnormal movements. Once trained, the model can quickly process new video data with high precision, reducing the time required for both training and inference while maintaining accurate biomarker determination.
Data Source
AI summary
A computer-implemented method includes obtaining a video of a subject, the video including a plurality of frames; generating, based on the plurality of frames, a plurality of optical flows; and encoding the plurality of optical flows using an autoencoder to obtain a movement-based biomarker value of the subject.