In-situ NIR spectroscopy study on microgravity-induced articular cartilage degeneration with ResNet-18

Articular cartilage (AC) plays a vital role in maintaining joint function, but it is highly vulnerable to degeneration due to the lack of mechanical loading under microgravity and hard to be detected in situ. Near-infrared (NIR) spectroscopy, known for its deep tissue penetration and rapid detection, has outstanding potential in cartilage assessment, but unexplored in evaluating cartilage degeneration under microgravity. In this study, a lab-made NIR probe was developed to acquire in situ NIR spectra of the control cartilage and the tail suspension (TS) cartilage for simulating microgravity 7, 14, and 21 days. The NIR spectral analysis indicated that regular main components (collagen and proteoglycans) fluctuation and degeneration of AC with the extending TS duration and the regional differences between different condyles. Then the NIR spectra were combined with support vector machine (SVM) and Residual Network (ResNet-18) models based on continuous wavelet transform for identifying the degeneration difference and multi-classifying. Binary classification between control and TS-21d samples achieved over 90 % accuracy with SVM or ResNet-18. In ternary classification task among the TS groups, ResNet-18 reached accuracies of 81.40 % (medial) and 81.82 % (lateral), which started to be better than SVM and demonstrated its superior robustness and generalization. In quaternary classification, only the ResNet-18 model was used achieving the accuracies of 87.76 % (medial) and 82.98 % (lateral) approaching the binary identification results. The experimental results demonstrate the potential of NIR spectroscopy with deep learning for in-situ cartilage degeneration study under different microgravity durations, offering a promising strategy for early diagnosis and fundamental research on joint tissue adaptation under aerospace microgravity.