Non-destructive assessment of tissue engineered cartilage maturity using visible and near infrared spectroscopy combined with machine learning

Tissue engineering is a promising approach to address the unmet clinical need for treating cartilage damage. Monitoring the characteristics of tissue-engineered cartilage constructs (TECs) during culture is critical for optimizing culture conditions and assessing TEC development. An important characteristic of TEC development is the ratio of glycosaminoglycans (GAGs) to DNA. However, current methods for quantifying GAGs and DNA are destructive, making them impractical for longitudinal monitoring. This study investigates the potential of optical spectroscopy in the visible and near-infrared (NIR) ranges coupled with machine learning (ML) for non-destructive prediction of GAGs and DNA in TECs. Six types of TECs were incubated for 7 or 28 days (n = 6 × 2 × 3). Spectral data were acquired from the constructs in the NIR and visible-short-NIR (VIS-SNIR) ranges using a fiber-optic probe and a multichannel spectrometer, followed by biochemical quantification of GAGs and DNA. ML models were developed for predicting the GAGs and DNA from the spectra and evaluated by leave-one-sample-out cross-validation after determining the optimal spectral range, preprocessing method, and ML algorithm. The performance of the models for predicting the construct's relative maturity (based on culture duration) using the predicted GAGs/DNA ratio was also evaluated. A NIR-based model was optimal for DNA content prediction (R2 = 0.81), while a VIS-SNIR-based model was optimal for GAGs prediction (R2 = 0.79). The predicted GAGs/DNA ratios were 100 % accurate in distinguishing immature TECs from more mature ones. Thus, optical spectroscopy combined with ML can enable non-destructive prediction of the maturity of TECs via predicting their DNA and GAG contents.