Analysis and Classification of 3-D Printed Collagen-Bioglass Matrices for Cellular Growth Utilizing Artificial Neural Networks
Date of Award
Master of Science (MS)
Biomedical and Chemical Engineering and Sciences
Michael B. Fenn
Christopher A. Bashur
Introduction: Recent advances have shown the influence of biomimetic design on modulation of the extracellular matrix (ECM) and the role stiffness of it plays in cellular behavior. 2-D models for cellular growth poorly recapitulate the natural environment of cells, whereas application of a 3-D ECM model allows for a more biomimetic design for modeling cellular behavior. Integration of 45S5 Bioglass provides and osteostimulatory mineral component to these 3-D models, allowing for better mimicry of orthopedic design, affecting the mechanical and physiochemical properties of the ECM and leading to an increase in osteogenic behavior. 3-D Bioprinting allows for custom fabrication of these matrices in specific geometric arrangements for a model closer to that of the natural environment. Characterization through Raman spectroscopy allows creation of a training and testing set for classification. These spectral datasets are noisy and high dimensional (features), and thus must be dimensionally reduced through feature selection or feature extraction. This study utilized manual selection, X² automated selection, principal component analysis (PCA) extraction, linear discriminant analysis (LDA) extraction, and combination PCA-LDA and LDA-PCA feature extraction methods. These features could then be fed into a type of artificial neural network (ANN), a feed-forward multilayer perceptron (MLP) for high performance classification. This formed the basis for future work involving 3-D bioprinting of biomimetic orthopedic biomaterial designs. Methods: Bovine-derived collagen type I was used to prepare collagen matrices (C-Gels). Bioglass was incorporated into 8mg/mL C-Gels (C-BGs). Genipin 0.0625% was used for chemical crosslinking and VA-086 1% w/v, with 11 mw/cm2 and 60s of exposure time was used for photo-crosslinking C-Gels (XC-Gels) and C-BGs (XC-BGs). Stiffer substrates were created using plastic compression and chemical crosslinking of 3.3mg/mL C-Gels (XPC). A375, VMM1, and Saos2 cells were grown atop 3.3 mg/mL C-Gels, XPC, and TCTP substrates, with A375 and Saos2 additionally seeded atop 10mg/mL C-Gels. The Saos-2 cell line was additionally seeded on 8mg/mL C-Gels, XC-Gels, 50% C-BGs, XC-BGs, and TCTP. The AlamarBlue cell viability assay was used to determine cellular proliferation at days 1, 3, 5, and 7. SEM imaging was performed after 7 days of growth on C-Gels, XC-Gels, 50% C-BGs, and XC-BGs then assessed for side of nodule-like structures using ImageJ. Raman spectral mapping was utilized to characterize substrates and create a training set for integration into an MLP ANN. Results and Conclusions: The AlamarBlue cell viability assay showed a significant increase in cellular proliferation in A375, VMM1, and Saos-2 cells seeded on 3.3mg/mL and 10mg/mL C-Gels, XPC, and TCTP, yet no significant change was observed when increasing collagen concentration on A375 cells and a significant reduction in cellular viability in Saos-2 cells. C-BGs decreased compressive modulus significantly versus C-Gels at 10% w/w, with no significant difference with 70% w/w Bioglass incorporation. Cellular proliferation was not significantly different in Saos-2 cells grown upon C-BGs and XC-BGs versus that of C-Gels, though XC-Gels had a significant increase to proliferation over C-Gels, C-BGs, and XC-BGs. SEM images of Saos-2 cells after 7 days of growth on C-Gels, XC-Gels, C-BGs, and XC-BGs were analyzed for size of nodule-like formations, indicating that there was a significant increase in size on C-BGs, more so with XC-BGs. However, the determination could not be made between ossified nodule formations and mineral agglomerates on cells. MLP ANNs with multivariate feature reduction showed a higher F1 model performance score over that of univariate and manual methods when classifying C-Gels, C-BGs, and Bioglass, with PCA-LDA having the highest; Manual: 44.62% X2 : 64.44% PCA: 74.76 % LDA: 97.87% LDA - PCA: 80.36% PCA - LDA: 98.58%.
Schmitt, Trevor, "Analysis and Classification of 3-D Printed Collagen-Bioglass Matrices for Cellular Growth Utilizing Artificial Neural Networks" (2018). Theses and Dissertations. 591.