Optimization of scanning parameters of digital breast tomosynthesis (DBT) is in general highly complex mainly because of the incompleteness of the imaging problem of DBT. There are many parameters that can affect the image quality of DBT including the view distribution, scan angle coverage, exposure distribution, and tube voltage to name a few. Image reconstruction algorithm may also play an important role in determining the image quality. Under a constraint of fixed radiation dose to the patient, it is highly desirable to optimize those parameters to achieve the highest diagnostic utility of DBT for a given detection task. A design-of-experiment (DOE) is a statistical method widely used for optimizing industrial fabrication processes that involve many parameters. Although its theoretical rigor has been questioned and challenged, its practical utility has attracted many investigators in their optimization problems of research and development. In this work, we applied the L18 DOE method to optimize 4 scanning parameters and reconstruction algorithm. Mass and micro-calcification detection task were considered, and a simulation and CIRS phantom were used for the experiment. We successfully optimized the scanning parameters and reconstruction algorithm to first order by use of the DOE method. Three levels of each scanning parameters were realized through an orthogonal array totaling 18 experiments to be conducted. Total-variation minimization algorithm outperformed the Filtered Back Projection (FBP) and Projection-Onto-Convex-Sets(POCS) algorithm, and optimum level of each scanning parameter was determined. In the simulation study, the contrast-to-noise ration of mass ranged from 2.16 to 174 revealing the near-optimum condition out of those 15 experiment conditions. In the real experiment study, the full width at half maximum of a micro-calcification ranged from 1.21 to 7.70 revealing the near-optimum condition out of those 16 experimental conditions. The main effect analysis suggested an optimum condition at which the contrast-to-noise ration of mass and full width at half maximum of the micro-calcification turns out to be the maximum. We have applied a DOE method in optimizing DBT scanning parameters for a given task, and successfully demonstrated its utility in such optimization problems. The method can be useful for many including the system developers and medical physicists.
유방암은 현대 여성 암 사망의 주요 요인이다. 기존의 맘모그래피를 통해서 얻을 수 있는 정보는 2D에 국한된 이미지이기 때문에 암 주위 조직에 의한 구조 노이즈로 인하여 암조직을 판별하기는 힘든 단점이 있다. 디지털 유방 단층 합성(Digital Breast Tomosynthesis)은 2D 이미지가 아닌 준3D 이미지를 의사에게 제공함으로써 암조직을 판별하기 용이하게 해준다. 하지만, 과다한 피폭 선량을 피하기 위해 CT처럼 360˚ 촬영을 하지 않고 제한된 스캔 각에서 제한된 뷰의 개수를 갖게 되어 있다. 스캔 조건에는 이와 더불어 노출량의 분포, 뷰의 분포 등 다양한 변수들이 존재하고, 이들을 최적화하는 것이 DBT의 성능을 결정하는 매우 중요한 부분이다. 본 연구에서는 다구찌 실험계획법을 이용하여 스캔조건의 최적화 연구를 하게 되었다. 다구찌 실험계획법은 생산 공정의 최적화와 같은 분야에 많이 사용되어 온 방법으로, 이를 이용하게 되면 많은 변수의 다양한 수준에 따른 필요 실험횟수를 크게 줄여주는 장점이 있다. 다구찌 실험계획법을 이용한 시뮬레이션 스터디 및 실제 Phantom을 이용하여 real data를 얻음으로써 다구찌 실험계획법이 신뢰할 만한 최적 조건을 도출해 낼 수 있음을 보였다.
