The effect of scaffold structure and biologic factors on bone regeneration in cap-based scaffolds measured using machine learning

Bone regeneration conduits are necessary in critically large defects. Hydroxyapatite (HAp) scaffolds are osteoconductive, synthetic grafts that provide an alternative to natural grafts, which have limitations. We and others have fabricated HAp scaffolds with a hierarchical structure that has macropores between orthogonal rods - into which bone can grow, and micropores within those rods. Our previous research has shown that these micropores create a capillary effect and, with the macropores, pull in fluid from the defect at implantation, encouraging bone regeneration. There is a limited understanding in the effect of different components of the defect fluid (cells and serum) and extracellular vesicles (EVs) on bone regeneration. In addition, the effect of relaxing HAp scaffold rod orthogonality has not been experimentally studied. This work aims to understand the effect of these structural and biological factors on bone regeneration in HAp scaffolds. We investigate the effect of scaffold structure (control rectilinear, optimized rectilinear, and optimized curvilinear), biologic treatments (control-dry, cells, serum, EVs), implantation site (anterior, middle, and posterior), and dry vs. wet implantation conditions (to further explore the capillary effect) on bone regeneration. Bone regeneration is measured using segmented micro-computed tomography (Micro-CT) images of implanted scaffolds, a challenging segmentation problem because of the similar mineral composition, and thus attenuation, between scaffold and ingrown bone. Convolutional U-Net models are increasingly used for segmentation of biomedical images as they do not require large training sets. Model segmentation performance is typically reported using aggregate metrics such as dice similarity coefficient (DSC), precision, and recall. These metrics, while useful, are reported as mean ± standard deviation without consideration of the context. As a result, these common metrics may not reflect the key results that are relevant to the application. Here, the key result is in accurately quantifying bone regeneration and in characterizing spatial variations in bone regeneration in HAp scaffolds. We implement a U-Net model trained with three loss functions—Cross-Entropy loss (widely used), Soft Dice loss (to address class imbalance) and Accuracy-based loss. We consider the parameter ΔBGF, the difference in bone growth fraction (BGF) between predicted output and ground truth images, and its correlations of the more standard metrics with |ΔBGF| for the models using the three loss functions. Accuracy is the most representative metric of performance relative to our outcome of interest, measuring ingrown bone. We show that the model is robust to changes in orientation, and resolution, and contrast – representing different imaging micro-CT conditions - supporting its potential for broader application. Using the results from the U-Net segmentation, we quantify bone regeneration in the scaffolds considering the treatments above and compare the outcomes in order to understand what factors contribute most to bone regeneration. The analysis considers both anatomic factors (implantation site (A, M, P), depth of scaffold in the defect (Hb/s)) and geometric factors created by the surgical procedures (defect to scaffold diameter ratio (∅d/s) – or the gap between scaffold and defect - and scaffold tilt (tan(θ))). The quantitative analysis shows that scaffolds implanted in the middle implantation site had significantly higher BGF than scaffolds in the anterior implantation site (p < 0.05). BGF variation is caused by other factors like bone contact fraction (BCF), defect to scaffold diameter ratio (∅d/s), the fraction of the scaffold height within the defect (Hb/s), and the tilt (tan(θ)) within the defect. When we isolate the BGF from BCF, ∅d/s, Hb/s, and tan(θ), we observe that all dry scaffolds (control rectilinear - dry, optimized rectilinear, optimized curvilinear) had significantly higher BGF than all wet scaffolds (cells, serum, EVs) (p < 0.05). This result is important because it supports our previous findings: fluid in the defect acts synergistically with these HAp scaffolds with multiscale porosity, through their capillarity, in enhancing bone regeneration [1,2]. The current study shows that this capillary effect is prominent after accounting for other factors like those caused by bone anatomy or geometry of the defect. The results further show that the combination of endogenous biologic components in the defect fluid is also synergistic compared to the individual components (cells, serum, EVs). All together, this study provides guidance for implant design at the scale of the defect and scaffold size and shape, as well as at the scale of the microstructural features of the scaffold that can generate the capillary forces to draw in fluid.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Aboutaleb, Sohaila

Contributors dc:contributor

• Wagoner Johnson, Amy
• Tawfick, Sameh
• Kersh, Mariana
• Norato, Julian

• Scaffolds
• regeneration
• U-Net
• segmentation

Statement dc:rights

• Copyright 2025 Sohaila Aboutaleb

Language dc:language

en

Handle dc:identifier

https://hdl.handle.net/2142/132659

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/132659