Seamless Scaffold-Free Hollow Constructs via Anchored Cell Sheet Engineering
Overview This computational and experimental pipeline supports the design, fabrication, and quantitative analysis of scaffold-free hollow tubular and spherical constructs biofabricated following Anchored Cell Sheet Engineering concept. This approach enables reproducible generation of mechanically stable, physiologically relevant hollow tissues for regenerative medicine. The platform provides tunable construct dimensions, supports multi-cell type architectures, and offers objective wall thickness quantification to evaluate biofabrication outcomes.
Features Curved-surface culture devices for single-step formation of hollow tubes and spheres without exogenous scaffolds and biomaterials. Multi-cell type layering to replicate complex in vivo-like tissue organization. Custom Python workflow for wall thickness measurement from histology whole slide images (WSIs). Quantitative control of wall thickness via core size and cell layering strategy.
Content
Step 1 — Parse SVGs from QuPath and build analysis metadata Whole-slide images (WSIs) prepared in QuPath® are exported as SVGs containing three layers: the base image, a contour line defining the region of interest (ROI), and a reference grid. The script parses each SVG, detects these layers by label (robust to minor naming differences), and rasterizes them at uniform resolution. For each sample, it extracts experimental metadata from the filename (Condition – Repeat), confirms canvas dimensions from the SVG viewBox/image tag, and writes a structured record to a metadata table. Each record contains traced file paths for the three layer renders, image dimensions, and a processing status flag. The resulting metadata.csv provides end-to-end traceability for subsequent steps.
Step 2 — Isolate the ring-only ROI and establish the pixel-to-micron scale Using the Path (contour) render, the script constructs a binary annular mask representing the vessel wall (inside the outer boundary minus the lumen). Gaps in the contour are closed via morphological operations, outer/inner regions are separated with connected-component logic, and small specks are removed. The mask is applied to the base image to produce a ring-only ROI image. To compute spatial scale, the script analyzes the Grid layer: adaptive thresholding and edge detection enhance the grid lines; probabilistic Hough transforms detect near-horizontal and near-vertical lines; clustered line positions yield a robust estimate of pixels per grid spacing. With the known grid spacing (1000 µm), the script derives a per-slide micron-per-pixel value. All outputs and the computed µm/pixel scalars are written back to metadata.csv.
Step 3 — Normal-based wall-thickness quantification and cohort-level visualization For each sample, a wall mask is rebuilt from the Path layer and the inner/outer boundaries are extracted. The inner boundary is resampled uniformly by arc length (default 360 samples per perimeter). At each resampled point, the local unit tangent is computed and rotated to obtain the outward normal. Wall thickness is measured by robust ray casting along the normal direction (with adaptive entry/exit confirmation and a small angular sweep if needed) to detect entry into and exit from the wall region. Distances are converted from pixels to micrometers using the per-slide scale from Step 2. Per-sample outputs include: (i) a CSV of per-degree thickness values and angles; (ii) a polar plot showing thickness as a function of perimeter angle; and (iii) a debug overlay depicting inner/outer boundaries and representative normal rays. Across samples, the pipeline generates cohort-level visualizations: a polar overlay of all samples (grouped by condition) and violin plots built from pooled per-degree measurements, with centered dots indicating each sample’s mean thickness. Statistical comparisons are run on two levels: (A) independent units (per-sample means; one value per sample) and (B) all individual points (pooled per-degree values). The pipeline performs one-way ANOVA and Tukey’s HSD when available; if post-hoc packages are unavailable, it falls back to Welch’s unequal-variance t-tests with Bonferroni correction. Results are saved as tidy CSV/TXT summaries alongside the figures.
Dependencies Hardware: Google Colab .
Software: Python 3.x, OpenCV, NumPy, Matplotlib, scikit-image, pandas, QuPath® (for ROI definition).
Contributions We welcome feedback, feature requests, and pull requests to extend the analysis pipeline and device design capabilities.
Credits This work was developed at Evolved.Bio as part of our mission to advance scaffold-free biofabrication for organ-scale regenerative medicine.
License This work is released under [insert link].
Contact For questions, collaborations, or licensing inquiries, contact: Alireza Shahin – alireza@itsevolved.com