Stereology neuron counts correlate with deep learning estimates in the human hippocampal subregions

Abstract

Hippocampal subregions differ in specialization and vulnerability to cell death. Neuron death and hippocampal atrophy have been a marker for the progression of Alzheimer's disease. Relatively few studies have examined neuronal loss in the human brain using stereology. We characterize an automated high-throughput deep learning pipeline to segment hippocampal pyramidal neurons, generate pyramidal neuron estimates within the human hippocampal subfields, and relate our results to stereology neuron counts. Based on seven cases and 168 partitions, we vet deep learning parameters to segment hippocampal pyramidal neurons from the background using the open-source CellPose algorithm, and show the automated removal of false-positive segmentations. There was no difference in Dice scores between neurons segmented by the deep learning pipeline and manual segmentations (Independent Samples t-Test: t(28) = 0.33, p = 0.742). Deep-learning neuron estimates strongly correlate with manual stereological counts per subregion (Spearman's correlation (n = 9): r(7) = 0.97, p < 0.001), and for each partition individually (Spearman's correlation (n = 168): r(166) = 0.90, p <0 .001). The high-throughput deep-learning pipeline provides validation to existing standards. This deep learning approach may benefit future studies in tracking baseline and resilient healthy aging to the earliest disease progression.

Figure 1

Cytoarchitecture characteristics of hippocampal subregions in Nissl staining: (a) medial (uncal) hippocampal subregions at the level of the hippocampal head: CA3u, CA2u, CA1u, Subu. (a’) Pyramidal layers of the medial hippocampal subregions. (b) Lateral hippocampal subregions at the level of the hippocampal body, clockwise: CA4, CA3, CA2, CA1, Sub (subiculum). (b’) Pyramidal layers of the lateral hippocampal subregions. Magnification bars in (a, b) = 1 mm; in (a’,b’) = 200 µm.

Figure 2

CellPose deep learning pipeline applied to hippocampal pyramidal neurons: (a) parcellated photomacrograph of hippocampus stained for Nissl substance (50 µm thick and coronal plane). (b) Cropped hippocampal pyramidal neuron layer to create partitions, transformed into inverted gray-value image. (c) Generating individual partitions. (d) Unfiltered segmented pyramidal neurons based on CellPose. Yellow outlines show neuron segmentations. (e) Filtered segmented neurons. Red X’s indicate removed items from segmentation (false-positives), (f) Hippocampal pyramidal neuron ellipsoid fitting and measurements (yellow: neuron segmentation, blue: fitted ellipse, red: neuron diameter (ellipsoid minor).

Figure 3

Manual hippocampal pyramidal neuron counts using optical fractionator stereology and systematic random sampling: (a) Nissl stained section with inset showing CA3 (example ROI). (b) ROI was outlined with random sampling grid overlaid and resulting with approximately 10 counting frames per the ROI. (c) 50 µm × 50 µm counting frame showing exclusion (red) and inclusion (green) lines. Tissue thickness taken by focusing from top to bottom of the section and averaged 20.3 µm across partitions. ROI region of interest.

Figure 4

Correlation of the CellPose deep learning pipeline: (a) Nissl stained vignette used for pyramidal neuron segmentation, three raters performed manual segmentations (masks 1, 2, 3; blue), and CellPose performed the automated segmentation (mask 4; brown). (b) No significant difference between the Dice scores calculated from the manual masks (manual rater versus manual rater), and automated masks (automated versus manual rater). Whiskers indicate min to max, + indicates mean, line indicates median. (c) Manual pyramidal neuron counts and automated pyramidal neuron estimates. Averaged across subregions, levels and cases. Whiskers indicate min to max, + indicates mean, line indicates median. (d) Subregion specific pyramidal layer neuron estimates, averaged across cases and levels. (e) Excellent correlation of manual (stereology) and automated (CellPose) pyramidal layer neuron estimates averaged per subregion (Spearman’s correlation (n = 9): r(7) = 0.97, p < 0.001). (f) Excellent correlation of manual (stereology) and automated (CellPose) pyramidal layer neuron estimates per partition (not averaged per subregion; Spearman’s correlation (n = 168): r(166) = 0.90, p < 0.001).