基於雙光子內視鏡與深度學習增強的高速深腦神經體積成像

Abstract

理解大腦的複雜結構與功能仍然是神經科學中的一項重大挑戰。雙光子激發顯微術克服了傳統光學顯微術的限制，實現了更深層的組織穿透、更低的光毒性，以及具備體積解析度的活體成像。近期的技術進展，例如可調式聲學梯度（TAG）透鏡，透過沿 z 軸快速調制焦點，進一步將雙光子激發拓展至高速體積成像。當 TAG 透鏡與梯度折射率（GRIN）透鏡結合時，可進行活體深層神經迴路成像，並達到近乎影像速率的體積獲取。我們開發了一套自製的雙光子內窺系統，結合 TAG 與 GRIN 透鏡，以捕捉位於小鼠大腦深層 6 毫米的視交叉上核（SCN）的神經動態。然而，這些技術進步的代價是在深層組織成像時影像對比度降低。為了提升影像品質，我們應用了基於深度學習的後處理方法，利用自監督模型挖掘時空冗餘以達到有效的降噪。此外，我們還開發了一套客製化演算法用於像差校正。我們的系統成功地獲取了活體中三維的神經元群體活動，並提升了影像的對比與清晰度，從而捕捉到 SCN 的神經動態。這個方法能幫助我們更清楚地研究晝夜節律和大腦深部的神經連結，也提供了一個實用的工具來進一步探索深腦中複雜的神經活動。

Understanding the intricate structure and function of the brain remains a critical challenge in neuroscience. Two-photon excitation microscopy has addressed the limitations of traditional optical microscopy, enabling deep tissue penetration, reduced phototoxicity, and volumetric resolution for in vivo imaging. Recent advancements, such as tunable acoustic gradient (TAG) lenses, have extended two-photon excitation capabilities to high-speed volumetric imaging by rapidly modulating focus along the z-axis. When integrated with gradient-index (GRIN) lenses, TAG lenses enable in vivo imaging of deep neural circuits, achieving nearly video-rate volumetric acquisition. We have developed a home-built two-photon endoscopy system incorporating TAG and GRIN lenses to capture neural dynamics in the suprachiasmatic nucleus (SCN), located 6 mm deep in the mouse brain. However, these advancements come at the cost of reduced image contrast, particularly during deep tissue imaging. To enhance image quality, we applied deep-learning-based post-processing techniques, leveraging self-supervised models that exploit spatiotemporal redundancies for effective denoising. Moreover, a customized algorithm is developed for aberration calibration. Our system successfully derived three-dimensional neuronal population activity in vivo with enhanced contrast and clarity, capturing neuronal dynamics in the SCN. This innovative approach provides a powerful tool for advancing our understanding of circadian rhythms and deep brain functional connectivity, paving the way for further exploration of complex neural processes.