AI ultrasound diagnosis machine

Artificial intelligence (AI)-based high-resolution ultrasound diagnosis machine

CR-3001C Ultrasound Imaging Diagnostic Device Design

CTU-3000 Ultrasound imaging is the latestArtificial Intelligence (AI) AlgorithmsThe latest development equipment that overcomes the technical limitations of existing equipment with differentiated technology

Algorithm performance verification video

Raw B-mode image

Applying the first algorithm

Applying the second algorithm

3rd algorithm applied

DL-SRU Imaging Algorithm for Blood Flow Velocity Field Measurement

An AI-equipped ultrasound imaging algorithm based on Deep Learning applying Deep-ULM and convolutional neural networks, which identifies the location of red blood cells (RBCs) in ultrasound images and reconstructs vascular images in high resolution using neural networks for synthetic tracking images without the need for contrast agents (CA).

Spatial resolution of image algorithms trained with synthetic data

Verification of excellent spatial resolution within a 1-pixel range at high resolution for various branching models after measuring vascular branching model images as average intensity images and training standard ULM and DL-SRU neural networks.
As tracer concentration increases, the accuracy of standard ULM decreases, but DL-SRU maintains high accuracy .
Ultrasound image acquisition time is 24ms for standard ULM and within 10ms for DL-SRU.
Standard ULM cannot locate the exact location of the RBC, but DL-SRU can track it within a 1-pixel radius.

Measurement using the 2-Frame PTV algorithm

RBCs are tracked by neural networks by training using RBC images synthesized under demanding biological conditions such as low frame rates, high concentrations, and organ movement.
Improvement of network robustness by training with various synthetic datasets using diverse hemocytosis, tracer density, scattering intensity, and SNR.
The tracer concentration in the synthetic image enables the P TV algorithm to be properly applied to the DL-SRU, thereby obtaining accurate RBC flow information.

DL-BDC Algorithm for Measuring Blood Flow and Blood Vessel Wall Behavior

Novel Deep-learning boundary detection and compensation DL-BDC technology in ultrasound imaging to accurately analyze hemodynamics and vessel wall interactions in arteries

Measurement accuracy dramatically improved through vector synthesis of wall and blood flow velocity in the wall-nearby region

Comparison of partitioning performance with other AI algorithms

Experimentally verified the performance of DeepLab, FCN, and Segnet AI algorithms, the performance of the developing USUNet, and the segmentation performance of the vessel interior and wall in vascular simulation models and animal vascular experiments.
Simulation verification with MATLAB and Python
Train other AI algorithms using the same training data set and hyperparameters used for USUNet training
Evaluate the segmentation performance of artificial neural networks by calculating DSC.
The partitioning performance of USUNet evaluated by DSC is very superior compared to other Deep_learning-based partitioning algorithms.

USUNet has been proven by achieving the best segmentation performance in tissue-mimicking phantoms and in vivo imaging.

Calculation of blood flow and velocity using continuous segment images after segmentation into the lumen and vessel wall
The velocity vector measured in the near-wall region is subtracted from the corresponding wall velocity vector component to compensate for the influence of the moving wall on flow dynamics.
Identifying other hemodynamic parameters by accurately calculating wall shear stress (WSS), vessel wall elasticity, and vibration shear number using corrected flow information
For multi-class segmentation, DL-BDC processes 400 ultrasound images
Using the proposed USUNet, the processing time required for segmenting the vessel wall and lumen can be reduced to less than 3ms.
The total processing time of DL-BDC for corrected flow and wall dynamics is possible within approximately 11.2 seconds per image pair.

An algorithm combining a hybrid method (AH) applying CDI and SIV

The accuracy of the Color Doppler Imaging (CDI) technique is enhanced by ensuring a sufficient angle of sound wave irradiation. Consequently, this can unintentionally compress perivascular tissue, potentially distorting actual vascular information. Furthermore, there are issues with measurement values varying depending on the operator, relatively low spatial resolution, and limitations in analyzing flow direction.
The SIV (Speckel Image Velocimetry) technique has lower temporal resolution than CDI, resulting in a relatively lower maximum measurable blood flow velocity, and there are limitations to the potential for performance improvement through enhancements in ultrasound imaging device hardware capabilities.
Applying an ultrasound imaging algorithm that combines CDI and SIV using an Adaptive Hybrid (AH) method significantly improves measurement accuracy without the need for additional hardware.
Significantly improving the accuracy of blood flow measurement by combining the high temporal resolution of the CDI technique and the excellent spatial resolution of the SIV technique

In the aging and DM groups, the endothelium becomes thinner and the blood vessel walls thicken compared to the control group.

Imaging of vascular inflammation in aging and DM groups

Carotid artery tissue analysis

Adaptive hybrid sequence

Color Doppler Image

High time resolution
Aliasing caused by Pulse Repettition Frequency
Single measurement direction, measurement accuracy varying with the angle of incidence of the ultrasound

Speckel Image Velocity

When applied, it provides higher spatial resolution than CDI and eliminates aliasing caused by phase shift.

Velocity field acquired by applying the SIV technique to B-MODE images and color Doppler images acquired at the same location
Identify errors in the CDI measurement vector based on the reference point and replace them with SIV measurement results.
The final velocity field result is derived by applying the AH method twice.

Measurement verification by accurately correcting measurement errors of the CDI method caused by low flow velocities or large velocity changes using the SIV method.

Apply the 1st Algorithm

Secondary Algorithm Application

3rd order algorithm applied

R&D System and Participating Organizations

This product was developed through industry-academia-research joint research with Core Teams, Pohang University of Science and Technology/Fluid Engineering Center, and Konyang University Hospital.

유사제품 성능 비교/ 제품 경쟁력

Products are compared based on information provided on the Internet, and there may be differences with the current equipment characteristics of each company.