Title

Real-Time Therapy Guidance Using Deep Learning- Assisted Focused Ultrasound Transducers

DGIST Authors

Yujeong shin ; Jin Ho Chang

Advisor

장진호

Issued Date

2026

Awarded Date

2026-02-01

Type

Thesis

Description

Focused Ultrasound (FUS), Deep-Learning, Barker Code, Deconvolution, Real-Time

Table Of Contents

Ⅰ. INTRODUCTION 1
1.1 Focused Ultrasound 1
1.2 Image Guidance and Monitoring 2
1.3 Integrated Therapy and Imaging for USgFUS 3
1.4 Limitations of Rule-Based Imaging for Linear Array FUS Transducers 5
1.5 CUDA Architecture for Real-Time System 6
1.6 Objectives of Research 7
ⅠI. METHODS 9
2.1 Overview of Proposed Framework 9
2.2 Rule-Based Imaging Pipeline and Ground Truth Generation 12
2.2.1 Channel-Level Ultrasound Signal Model 12
2.2.2 Wiener Deconvolution for Axial Resolution Enhancement 13
2.2.3 Barker Code Compression Using PSL Mismatched Filter 14
2.2.4 Band-Pass Filtering and Normalization 15
2.2.5 Ground Truth Generation Procedure 15
2.3 Deep Learning Models for Scanline-Level RF Enhancement 16
2.3.1 Problem Formulation 16
2.3.2 Network Architecture 18
2.3.3 Training Strategy and Loss Function 18
2.4 CUDA-Based Real-Time System Architecture 19
2.4.1 GPU-Based Processing Pipeline 20
2.4.2 Deep-Learning Inference with TensorRT 21
2.4.3 Multi-Stream Execution for Real-Time Processing 22
ⅠII. RESULTS 24
3.1 Comparison of Deep Learning Models 24
3.2 Experimental Setup and Data Acquisition for Imaging Experiments 26
3.3 Quantitative Image Quality Evaluation 30
3.3.1 Evaluation on Axial Resolution 30
3.3.2 Evaluation on CNR and SNR 32
3.3.3 Evaluation on Ex-Vivo 33
3.4 Evaluation on Computational Efficiency and Frame Rate 36
3.4.1 Comparison between Rule-Based and Deep Learning Pipelines 36
3.4.2 Effect of Multi-Stream Processing on Frame Rate 37
IV. DISCUSSION AND CONCLUSION 38
REFERENCES 41

URI

https://scholar.dgist.ac.kr/handle/20.500.11750/59731
http://dgist.dcollection.net/common/orgView/200000947874

DOI

10.22677/THESIS.200000947874

Degree

Master

Department

Artificial Intelligence Major

Publisher

DGIST

Show Full Item Record