Ultrasonic Probe Stitching via Neural Network Motion Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image stitching technologies for ultrasonic probes in medical applications suffer from low accuracy and high system costs due to the limitations of conventional registration methods and the need for expensive electromagnetic positioning systems.
Innovation Solution
A method using a pre-trained neural network to calculate a transformation matrix based on motion data from sensors, such as accelerometers and gyroscopes, to stitch images without relying on image features, thereby improving accuracy and reducing system costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional registration technology is used to stitch ultrasonic images, then panoramic image can be obtained, but stitching accuracy is low
Solution Approach 1:
The patent replaces the conventional image feature-based registration method with a sensor-based motion tracking system. Accelerometers and gyroscopes mounted on the ultrasonic probe directly measure the probe's motion, and neural networks process this sensor data to calculate transformation matrices for image stitching. This substitution of mechanical sensing for image processing significantly improves stitching accuracy and reliability.
2Measurement precision
If electromagnetic positioning system is added to improve stitching accuracy, then panoramic image quality improves, but system cost increases significantly
Solution Approach 1:
The patent uses inexpensive accelerometer and gyroscope sensors instead of expensive electromagnetic positioning systems. These low-cost inertial sensors provide sufficient motion tracking data for accurate image stitching, dramatically reducing system cost while maintaining high stitching accuracy. The neural network processing further enhances the value extracted from this cheap sensor data.
Solution Approach 2:
The patent introduces neural networks as an intermediary between the sensor data and the transformation matrix calculation. The neural networks process the raw accelerometer and gyroscope data to derive accurate probe motion trajectories, enabling high-precision stitching without requiring expensive electromagnetic positioning hardware. This intermediary processing layer maximizes the utility of low-cost sensors.
3Ease of operation
If image feature-based registration is used, then stitching can be performed, but the method is complex and computationally intensive
Solution Approach 1:
The patent extracts the motion information from the image processing task and places it in the sensor system. Instead of analyzing image features to determine probe motion, the accelerometers and gyroscopes directly measure the motion, and neural networks process this extracted motion data to generate transformation matrices. This extraction simplifies the overall system by separating motion measurement from image processing.
Data Source
AI summary
The present disclosure discloses a panoramic stitching method, an apparatus, and a storage medium. A transformation matrix obtaining method includes: obtaining motion data detected by sensors, wherein the sensors are disposed on a probe used to collect images, and the motion data is used to represent a moving trend of the probe during image collection; inputting the motion data into a pre-trained neural network, to calculate matrix parameters by using the neural network; calculating a transformation matrix by using the matrix parameters, wherein the transformation matrix is used to stitch images collected by the probe, to obtain a panoramic image. In the present disclosure, the transformation matrix can be calculated and the images can be stitched without using characteristics of the images, and factors such as brightness and the characteristics of the images do not impose an impact, thereby improving transformation matrix calculation accuracy, and improving an image stitching effect.