Enhanced multiscale human brain imaging by semi-supervised digital staining and serial sectioning optical coherence tomography

Abstract

A major challenge in neuroscience is visualizing the structure of the human brain at different scales. Traditional histology reveals micro- and meso-scale brain features but suffers from staining variability, tissue damage, and distortion, which impedes accurate 3D reconstructions. The emerging label-free serial sectioning optical coherence tomography (S-OCT) technique offers uniform 3D imaging capability across samples but has poor histological interpretability despite its sensitivity to cortical features. Here, we present a novel 3D imaging framework that combines S-OCT with a deep-learning digital staining (DS) model. This enhanced imaging modality integrates high-throughput 3D imaging, low sample variability and high interpretability, making it suitable for 3D histology studies. We develop a novel semi-supervised learning technique to facilitate DS model training on weakly paired images for translating S-OCT to Gallyas silver staining. We demonstrate DS on various human cerebral cortex samples, achieving consistent staining quality and enhancing contrast across cortical layer boundaries. Additionally, we show that DS preserves geometry in 3D on cubic-centimeter tissue blocks, allowing for visualization of meso-scale vessel networks in the white matter. We believe that our technique has the potential for high-throughput, multiscale imaging of brain tissues and may facilitate studies of brain structures.

Fig. 1. Overview of the proposed OCT DS technique.

a Data acquisition and DS model. S-OCT alternates between imaging and tissue sectioning to acquire a stack of block-face OCT images, which are then processed to compute the scattering coefficient (OCT-SC) map stack. Sectioned sample slices are physically stained and imaged. The DS neural network is trained from a few weakly-aligned pairs of OCT-SC and Gallyas silver-stained images. b After the DS model is trained, it can perform inference on completely new slices of OCT-SC images for volumetric DS

Fig. 2. The training framework of our DS neural network model.

a The backbone of the DS network G is built on the CUT framework, which combines contrastive learning and adversarial learning. The input is a 2D OCT-SC map X and the output is a digitally stained image G(X) that is compared with a PS image Y from an adjacent slice. b Auxiliary pseudo-supervised learning task. The biophysical module computes the optical density OD(Y) of the PS image Y, which is fed as an input to G. The digitally stained OD image G(OD(Y)) is compared with the original PS image Y during training. c Auxiliary unsupervised cross-modality image registration task. We alternate between optimizing G and a registration network R under different image scales. We fix R while updating G, which provides more informative supervision for R in the next iteration. We use patch-wise losses for training G, and whole slide image (WSI) losses for training R. The red and green channels of the deformation field represent the vertical and horizontal displacement vectors, respectively

Fig. 3. DS results on OCT-SC of tissue slices and comparisons with PS images.

Cases include (a) ideally stained slices; (b) non-uniformly stained and under-stained slices. ROI 1, 3, 5, and 7 are gyral crest regions and 2, 4, 6, and 8 are sulcal fundus regions. VS: “vessel space”. Scale bars are 1 mm. The second slice in (a) is an independent testing slice from a subject that also contributed other slices in the training set, while all other shown slices are from entirely independent testing subjects

Fig. 4. Comparisons results of layer differentiation and thickness estimation in DS results.

a The DS and PS WSIs from a cortex tissue section. b Zoom-in ROIs of inverted OCT-SC, DS and PS modalities marked in green and red boxes in (a) and normalized intensity profiles aggregates along white dotted lines. c Manually annotated layers IV/V/VI labeled in three colors and estimated local thickness. Statistics of thickness are visualized in box plot and grouped by gyral crest and sulcus regions. ROI is the zoom-in of the dotted blue box from (a). The shown slice is an independent testing slice from a subject that also contributed other slices in the training set

Fig. 5. 3D visualization and cross-section views of the DS results on a large unseen tissue block.

a The DS output images are stacked along the z-axis to render the whole digitally stained volume as well as segmented white matter regions. b Orthogonal cross-sectional views of the DS volume. c Two zoom-in regions of vessel structures in yellow and green boxes from (a) are shown on the left. Three orthogonal maximum intensity projections (MIP) of the DS volume are shown on the right. All scale bars are 5 mm. The shown sample comes from an unseen subject entirely independent of training and testing subject set

Fig. 6. DS-OCT generalization performance on an unseen hippocampus tissue slice.

Examples of OCT-SC, DS, and PS images (of adjacent sections) on one sample from the Hippocampus region are shown. SP: Stratum Pyramidale; AL: Alveus; FF: Fimbria Fomix; SR: Stratum Radiatum; SM: Stratum Moleculare; DG: Dentate Gyrus; CA: Cornu Ammonis. The shown slice comes from an unseen subject entirely independent of training and testing subject set