Development of high-resolution multispectral fluorescence lifetime imaging microscopy for characterization of atherosclerotic plaque

Abstract

죽상 동맥경화는 동맥 벽에 플라크 형성으로 특징 지어지며, 지질과 거품 세포의 축적으로 인한 염증반응으로 발달된다. 따라서 동맥경화의 발달 및 진단을 위해서는 동맥의 조직 구성을 분석하는 것이 중요하다. 형광 수명은 형광 신호의 감쇠율로 정의된다. 이 형광 수명은 생체 내의 조직의 구성성분에 따라 서로 다르다는 특성을 갖고 있다. 특히 그 구성성분 중에 콜라젠, 엘라스틴, NADH와 같은 성분은 자외선의 여기 하에서 강한 자가 형광을 띄고 있다. 이 성질을 이용하여 자가 형광 통해 형광 수명 얻는 현미경을 설계하여 영상화 하면, 부가적인 조영제를 사용하지 않고 동맥경화를 구성하는 조직의 생화학적 정보를 얻을 수 있다. 이전 연구에서 형광 수명을 이용하여 조직의 구성성분을 영상화 하고 분석하는 연구를 진행하였지만, 분해능의 한계, 단일 채널의 신호 획득으로 인한 정보의 한계성으로 많은 어려움이 있었다. 본 연구의 목적은 고해상도의 다중 스펙트럼의 형광 수명 현미경을 개발하고 이를 이용하여 동맥경화 관상동맥 혈관을 이미징, 분석하는 것이다. 이 형광 수명 현미경은 광학 시스템 설계 프로그램을 이용하여 설계 및 검증 되었다. 여러가지 시뮬레이션을 통해 모든 파장에서 광학적 오차가 최소화 되게끔 설계하였다. 영상 획득은 멀티 스레드 형식의 소프트웨어를 개발하여, 실시간 획득이 가능하게끔 하였고, 여러 장비 제어가 가능하게끔 하였다. 개발한 시스템을 형광 용액을 이용하여 calibration을 진행하였고, USAF target을 이용하여 개발한 시스템의 해상도를, 형광 기준 용액을 이용하여 형광 수명 획득의 안정성을 평가하였다. 검증된 장비로 실제 돼지의 동맥경화 유발된 혈관의 단면을 영상화 해보았고, 세기 신호로 구분되지 않는 혈관의 내막들이 형광 수명으로 잘 구분됨을 확인하였다. 또한, 이 세기 신호가 고려된 형광 수명 영상을 얻어서, 좀 더 명확하게 생화학적 정보를 유추할 수 있게 하였다. 2 축의 전동 스테이지를 이용하여 여러 장의 이미지를 얻었고, 이를 이어 붙여서 실제 큰 부분의 이미지를 획득하였다. 획득한 이미지와 조직 병리학적 염색을 진행한 이미지를 비교하여 형광 수명 정보가 각 조직 구성성분과 어떠한 연관성을 보이는 지 알 수 있음을 확인하였다. 형광 수명 현미경으로 얻을 수 있는 값들 중 1 채널의 형광 수명, 2 채널의 형광 수명, 채널 1과 2의 세기 신호의 비율을 형광 수명 특징 변수로 정의하고 이와 조직 구성 성분과의 연관성을 보기 위해 통계적 분석을 진행하였다. 분석을 진행하기에 앞서 염색된 정도를 색상 정보를 바탕으로 뽑아 내었고, 관심 영역을 설정하는 알고리즘을 개발하여 한 장의 이미지 당 144개의 관심 영역을 얻었다. 획득한 관심 영역을 통해 단순선형회귀분석을 진행한 결과, 지방질과 형광 수명 간에는 음의 상관관계, 대식 세포와 형광 수명 간에 음의 상관관계 그리고 콜라젠과 형광 수명 간에 양의 상관관계를 갖고 있음을 통계적으로 유의미하게 확인하였다(p < 0.0001). 단순선형회귀분석은 구성 성분 간의 관계를 고려하기 힘들다는 단점이 있기에, 다중선형회귀분석을 설계하여 진행하였다. 이를 토대로 지방질과 형광 수명은 음의 상관관계, 콜라젠과 형광 수명은 양의 상관관계를 갖고 있지만, 대식 세포와 형광 수명 간에는 상관 관계가 없음을 확인하였다 (p < 0.0001). 따라서, 실제 형광 수명을 감소시키는 요인은 지방질이며, 대식 세포는 지방질과 큰 연관성이 있지만, 실제로 형광 수명에는 기여하지 않음을 알 수 있었다. 또한, 염색된 부분의 형광 수명 값을 확률밀도함수로 그려보아, 거품 세포의 형광 수명의 분포를 확인할 수 있었다. 앞서 설정한 형광 수명 특징 변수를 통해 각 조직 구성성분의 양을 예측하는 다중회귀분석모델과 인공신경망을 설계하였고, 통계적으로 유의미하게 잘 예측되었음을 확인하고, 실제 콜라젠의 양을 유추해 봄을 통해 예측 모델을 검증하였다. 이 연구에서 제안한 고해상도 다중 스펙트럼의 형광 수명 현미경을 통해 동맥경화 혈관을 영상화하고, 이를 조직 병리학적 염색 영상과 매칭, 통계적 분석을 통해서 형광 수명과 조직 구성 성분 간의 관계를 알 수 있었으며, 형광 수명 특징 변수를 통해 조직 구성 성분을 유추할 수 있었다. 향후 적대 신경망 같은 딥러닝 알고리즘을 활용하면, 가상 조직 검사 이미지도 구현할 수 있을 것이라 기대되며, 이를 통해, 심혈관계의 조직 관련 연구 뿐만 아니라 종양학 같은 형광 수명 영상이 가능한 분야에서도 활용하여 조직학적 연구에 대한 전반적인 이해나 연구를 촉진할 수 있을 것이다.|Atherosclerosis is characterized by the formation of plaque in the artery walls and develops as an inflammatory reaction due to the accumulation of lipid and foam cells. Therefore, it is important to analyze the arterial tissue components for the development and diagnosis of atherosclerosis. Fluorescence lifetime is defined as the decay rate of the fluorescent signal. And, fluorescence lifetime has characteristics that different types of tissue components have a different lifetime. In particular, components such as collagen, elastin, and NADH have strong autofluorescence under UV excitation. With these characteristics, biochemical information of atherosclerotic tissue components can be obtained without any contrast agent by using fluorescence lifetime imaging microscopy using autofluorescence. In the previous study, imaging and analysis of tissue compositions using fluorescence lifetime were difficult due to the limit of resolution and information from the only single-channel acquisition. The purpose of this study is the development of a high-resolution multispectral fluorescence lifetime imaging microscopy to analyze atherosclerotic coronary arteries.

With optical simulation, design and validation were being held about this fluorescence lifetime imaging microscopy. Various simulations are designed to minimize optical aberrations at all three channels. With multi-thread structured software, it makes enables imaging acquisition in real-time and control multiple devices. Calibration was being held with a fluorescence reference solution, and the resolution of the system was evaluated with a USAF resolution target. A commercial fluorescence slide and fluorescent solution were used to validate the stability in the acquisition of fluorescence lifetime.

After validation, slices that cross-sectioned the atherosclerotic coronary artery of the swine model were imaged. With fluorescence lifetime, it can be confirmed that the layers of coronary arteries had been well distinguished while intensity images cannot. Besides, intensity weighted fluorescence lifetime images were obtained to allow us to infer biochemical information more clearly. Multiple images were obtained with 2-axis motorized stage and large field of view imaging was made possible with stitching methods. By comparing the acquired images with histopathological staining images, it was confirmed that the fluorescence lifetime correlated with each tissue component.

Statistical analysis was being held with FLIM parameters, which is defined as channel 1 lifetime, channel 2 lifetime and intensity ratio of channel 2 and channel 1, to confirm correlation with the tissue components. Before proceeding with the analysis, the stained percentage was extracted based on the color information, and the algorithm for setting the region of interest was developed to obtain 144 ROIs per single image. With single linear regression analysis using ROIs, lipid and macrophage were confirmed to have a negative correlation with fluorescence lifetime, while collagen has a positive correlation with fluorescence lifetime, and it was confirmed to have statistically meaningful information (p < 0.0001). Since single linear regression analysis does not consider the correlation between biochemical components, multiple linear regression model was designed and held for analysis. With this analysis, lipid and fluorescence lifetime were seemed to have a negative correlation while collagen has a positive correlation with fluorescence lifetime. However, it is confirmed that there was no correlation between macrophages and fluorescence lifetime (p < 0.0001). Thus, it can be inferred that the factor that decreases the fluorescence lifetime is a lipid, and macrophages are highly correlated with lipid but do not contribute to fluorescence lifetime. Also, by plotting the probability density function of fluorescence lifetime for the stained region, the distribution of fluorescence lifetime for foam cells was confirmed. Multiple linear regression analysis models and artificial neural networks were designed to predict the amount of each tissue component using FLIM parameters with confirming statistically. The prediction models were verified by inferring the actual amount of collagen content.

In this study, the high-resolution multispectral fluorescence lifetime imaging microscopy was proposed to image atherosclerotic vessels and matched with histopathological staining images. Some statistical analysis was being held to reveal the relationship between fluorescence lifetime and tissues components and with estimating model, some biochemical tissue components were well estimated. In the future, using deep learning algorithms such as generative adversarial networks, it is expected that virtual histology images can be implemented. Through this, general understanding or research can be promoted in the fields of not only cardiovascular tissue research but also other fields, which fluorescence lifetime imaging is possible, such as oncology.; Atherosclerosis is characterized by the formation of plaque in the artery walls and develops as an inflammatory reaction due to the accumulation of lipid and foam cells. Therefore, it is important to analyze the arterial tissue components for the development and diagnosis of atherosclerosis. Fluorescence lifetime is defined as the decay rate of the fluorescent signal. And, fluorescence lifetime has characteristics that different types of tissue components have a different lifetime. In particular, components such as collagen, elastin, and NADH have strong autofluorescence under UV excitation. With these characteristics, biochemical information of atherosclerotic tissue components can be obtained without any contrast agent by using fluorescence lifetime imaging microscopy using autofluorescence. In the previous study, imaging and analysis of tissue compositions using fluorescence lifetime were difficult due to the limit of resolution and information from the only single-channel acquisition. The purpose of this study is the development of a high-resolution multispectral fluorescence lifetime imaging microscopy to analyze atherosclerotic coronary arteries.

With optical simulation, design and validation were being held about this fluorescence lifetime imaging microscopy. Various simulations are designed to minimize optical aberrations at all three channels. With multi-thread structured software, it makes enables imaging acquisition in real-time and control multiple devices. Calibration was being held with a fluorescence reference solution, and the resolution of the system was evaluated with a USAF resolution target. A commercial fluorescence slide and fluorescent solution were used to validate the stability in the acquisition of fluorescence lifetime.

After validation, slices that cross-sectioned the atherosclerotic coronary artery of the swine model were imaged. With fluorescence lifetime, it can be confirmed that the layers of coronary arteries had been well distinguished while intensity images cannot. Besides, intensity weighted fluorescence lifetime images were obtained to allow us to infer biochemical information more clearly. Multiple images were obtained with 2-axis motorized stage and large field of view imaging was made possible with stitching methods. By comparing the acquired images with histopathological staining images, it was confirmed that the fluorescence lifetime correlated with each tissue component.

Statistical analysis was being held with FLIM parameters, which is defined as channel 1 lifetime, channel 2 lifetime and intensity ratio of channel 2 and channel 1, to confirm correlation with the tissue components. Before proceeding with the analysis, the stained percentage was extracted based on the color information, and the algorithm for setting the region of interest was developed to obtain 144 ROIs per single image. With single linear regression analysis using ROIs, lipid and macrophage were confirmed to have a negative correlation with fluorescence lifetime, while collagen has a positive correlation with fluorescence lifetime, and it was confirmed to have statistically meaningful information (p < 0.0001). Since single linear regression analysis does not consider the correlation between biochemical components, multiple linear regression model was designed and held for analysis. With this analysis, lipid and fluorescence lifetime were seemed to have a negative correlation while collagen has a positive correlation with fluorescence lifetime. However, it is confirmed that there was no correlation between macrophages and fluorescence lifetime (p < 0.0001). Thus, it can be inferred that the factor that decreases the fluorescence lifetime is a lipid, and macrophages are highly correlated with lipid but do not contribute to fluorescence lifetime. Also, by plotting the probability density function of fluorescence lifetime for the stained region, the distribution of fluorescence lifetime for foam cells was confirmed. Multiple linear regression analysis models and artificial neural networks were designed to predict the amount of each tissue component using FLIM parameters with confirming statistically. The prediction models were verified by inferring the actual amount of collagen content.

In this study, the high-resolution multispectral fluorescence lifetime imaging microscopy was proposed to image atherosclerotic vessels and matched with histopathological staining images. Some statistical analysis was being held to reveal the relationship between fluorescence lifetime and tissues components and with estimating model, some biochemical tissue components were well estimated. In the future, using deep learning algorithms such as generative adversarial networks, it is expected that virtual histology images can be implemented. Through this, general understanding or research can be promoted in the fields of not only cardiovascular tissue research but also other fields, which fluorescence lifetime imaging is possible, such as oncology.