NeRF를 활용한 CT 재구성 시스템 설계 및 구현 = Design and Implementation of Computed Tomography Reconstruction System Using Neural Radiance Fields Techniques

국문 초록 (Abstract)

현대 의료 분야에서의 수술은 높은 수준의 정밀성과 안전성을 요구하는 복잡한 과정으로 인정받고 있다. 그러나 이러한 수술 프로세스에서 발생할 수 있는 방사능 문제는 여전히 중요한 과제로 남아 있다. 환자와 의료진의 안전을 최우선으로 고려할 때, 노출되는 방사능의 양을 최소화하고자 하는 노력은 계속되고 있으나 부가적인 장비 착용 등 부수적 문제가 존재한다. 이를 해결하고자, 본 논문에서는 방사능 노출과 관련된 문제를 해결하기 위한 혁신적인 소프트웨어적 방법을 제안한다. NeRF(Neural Radiance Field)와 새로운 손실함수를 도입하여, 부분 다 시점 2D 디지털화재구성사진(DRR)을 3D CT로 재구성하는 효과적인 방법을 제시한다. 이 방법은 방사능 투과 수식을 모델내에 암시적 학습하는 새로운 접근 방식을 채택한다. 머리, 상복부, 발데이터를 통하여 해당 방법을 정량적, 정성적으로 평가하였다. 해당 논문을 통한 3D 재구성 방법이 의료 분야에서의 안전성과 효율성을 향상시키는 새로운 기술적 해결책으로 기여할 것으로 예상된다.

다국어 초록 (Multilingual Abstract)

Surgery in modern medicine is renowned for its meticulous precision and emphasis on safety, requiring intricate procedures. However, ongoing concerns about radiation exposure during these surgical procedures remain a significant challenge. Continued efforts are being made to minimize radiation exposure, primarily focusing on ensuring the safety of patients and medical personnel. However, these efforts face additional hurdles, including wearing additional protective equipment. Given these challenges, this paper presents a pioneering software-based approach to mitigate radiation exposure issues. The proposed methodology introduces the Neural Radiance Field (NeRF) and a novel loss function to facilitate the reconstruction of three-dimensional CT scans from partial multi-view two-dimensional digitally reconstructed radiographs (DRRs). By employing a novel approach that implicitly learns the radiation attenuation equations within the model, this method aims to provide a comprehensive solution. The proposed approach's evaluation includes quantitative and qualitative assessments using head, abdomen, and feet datasets. The 3D reconstruction method outlined in this thesis is anticipated to serve as a transformative technological intervention, promoting improvements in safety protocols and operational efficiencies within the medical field.

참고문헌 (Reference)

• 1 "https://www.bionity.com/en/encyclopedia/Digitally_Reconstructed_Radiograph.html"
• 2 "http://www.monews.co.kr/news/articleView.html?idxno=62609"
• 3 D. F. Swinehart, "The beer-lambert law" 39 (39): 327-380, 1962
• 4 A. Biguri, "TIGRE : a MATLAB-GPU toolbox for CBCT image reconstruction" 2 (2): 055010-, 2016
• 5 A. Yang, "Research on feature extraction of tumor image based on convolutional neural network" 7 : 24204-24213, 2019
• 6 Z. Liu, "Recent progress in transformer-based medical image analysis" 164 : 107268-, 2023
• 7 Haykin, Simon, "Neural networks: a comprehensive foundation" Prentice Hall PTR 1994
• 8 B. Mildenhall, "Nerf : Representing scenes as neural radiance fields for view synthesis" 65 (65): 99-106, 2021
• 9 J. T. Barron, "Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields" IEEE 2021
• 10 T. Müller, "Instant neural graphics primitives with a multiresolution hash encoding" 41 (41): 1-15, 2022
• 11 V. Sitzmann, "Implicit neural representations with periodic activation functions" 33 : 7462-7473, 2020
• 12 M. Tancik, "Fourier features let networks learn high frequency functions in low dimensional domains" 2020
• 13 S. Ruder, "An overview of gradient descent optimization algorithms"
• 14 D. P. Kingma, "Adam: A method for stochastic optimization"
• 15 B. Kerbl, "3D Gaussian Splatting for Real-Time Radiance Field Rendering" 42 (42): 1-14, 2023