Title

Multi-view Training with SAM2 for Semi-Supervised 3D ABUS Image Segmentation

Alternative Title

SAM2를 활용한 반지도 3D ABUS 영상 분할을 위한 다중 뷰 학습

DGIST Authors

Dong Min Lee ; Sang Hyun Park ; Sung Hoon Im

Advisor

박상현

Co-Advisor(s)

Sung Hoon Im

Issued Date

2025

Awarded Date

2025-02-01

Citation

Dong Min Lee. (2025). Multi-view Training with SAM2 for Semi-Supervised 3D ABUS Image Segmentation. doi: 10.22677/THESIS.200000841604

Type

Thesis

Description

Automated Breast Ultrasound (ABUS), Semi-Supervised Learning, Segment Anything Model

Table Of Contents

I. INTRODUCTION 1
II. RELATED WORKS 3
1 Semi-supervised medical Image Segmentation 3
2 Segment Anything Models and Adaptation 3
3 Semi-Supervised Medical Image Segmentation With SAM 4
III. Method 5
1 Problem Setting 5
2 Overview of The Proposed Method 5
3 Supervised Adaptation of SAM2 with Detection Token 6
4 Detection Loss 7
5 Semi-Supervised Learning of SAM2 with Multi-Axis Consistency 8
IV. RESULTS 10
1 Experimental Setting 10
2 Results 11
2.1 Comparison with State-of-The-Art Methods 11
2.2 Ablation studies 12
V. CONCLUSION 15
References 16

URI

http://hdl.handle.net/20.500.11750/58103
http://dgist.dcollection.net/common/orgView/200000841604

DOI

10.22677/THESIS.200000841604

Degree

Master

Department

Artificial Intelligence Major

Publisher

DGIST

Show Full Item Record