Abstract Osteoporosis is a widespread chronic metabolic bone disease. It has developed into a severe public health problem, especially for older people. Preventing fractures caused by osteoporosis and enhancing patient experiences require early identification of bone mineral loss. The commercially available technologies for diagnosing osteoporosis are costly and may trigger cell mutation. Therefore, non-invasive treatments have received great attention. This work explores the potential of employing optical collimated transmittance at 660, 785, and 980 nm to monitor progressive mineral loss in EDTA-treated chicken bone samples. Over the course of six weeks, 20 samples (10 diaphysis and 10 epiphysis) were measured weekly. Transmittance was elevated following demineralisation, particularly at 980 nm, which increased by 190%. The receiver operating characteristic (ROC) analysis produced area under the curve (AUC) values ranging from 0.834 to 0.996 (95% CI), demonstrating strong discrimination between baseline and demineralised states. These results show that near-infrared optical transmission is sensitive to mineral loss and has the potential to be used for low-cost bone assessments. Although EDTA-induced demineralisation does not fully recreate the complicated pathophysiology of clinical osteoporosis, this simplified model serves as a preliminary platform for assessing optical signals associated with mineral loss.

Abstract Manual evaluation and interpretation of Impact echo (IE) data is often labor-intensive and time-consuming, motivating the growing interest in applying machine learning (ML) techniques to this non-destructive testing (NDT) method. However, the scarcity of labeled datasets limits the generalizability of ML models to new, unseen data. This study investigates strategies to integrate a-priori knowledge into Convolutional Neural Networks (CNNs) for improved prediction of the S1 Lamb wave frequency from IE signals. To this end, time–frequency representations of IE signals, derived using the Short-Time Fourier Transform (STFT), are used as model inputs. A-priori knowledge is introduced in the form of initial frequency estimates obtained through manual evaluation. Additionally, transfer learning is employed to enrich the limited measurement dataset with data from 2D numerical simulations. The results demonstrate that, although training loss curves remain similar across models, incorporating additional information significantly enhances performance on unseen datasets. Furthermore, pre-training with simulation data accelerates convergence during early fine-tuning stages. The highest predictive accuracy was achieved when the initial guess was directly embedded into the loss function.