|
The kidney is an important organ for excreting metabolic wastes and maintaining the stability of the internal environment in the body. The renal tubule is an essential structure for the nephron with reabsorption and excretion. Structural changes in the renal tubules can lead to kidney dysfunction and thus cause renal diseases. With the development of imaging technology, mesoscopic optical imaging can obtain kidney images at cell resolution. It is significant to reconstruct the three-dimensional (3D) morphology of renal tubules from the image to understand renal function and explore the pathogenesis of renal diseases. However, the large volume of high-resolution image data and the extensive spatial distribution of the renal tubule throughout the kidney present significant challenges for 3D reconstruction. To address this, we propose a deep learning-based method for renal tubule reconstruction. First, we propose a deep learning-based method for renal tubule reconstruction. First, we imaged mouse kidneys using High-Definition Fluorescent Micro-Optical Sectioning Tomography (HD-fMOST) to obtain kidney images at cellular resolution. We then employed a U-Net model to segment the renal tubules in two-dimensional (2D) kidney images, producing binary segmentation results. Finally, we performed the connected domain analysis of the segmentation results in 3D space and reconstructed the 3D morphology of all renal tubules. Our method demonstrates efficient and accurate reconstruction of renal tubules in mesoscopic kidney images.
|
