A deep learning enabled in-silico animal study approach for bone tissue engineering

Engineered tissue constructs are typically evaluated by extensive animal studies to assess their in vivo efficacy. Histological analysis is the widely used approach. However, this technique is time-consuming and highly dependent on experienced pathologists. Moreover, it provides limited insight into how tissue continuously changes at the same tissue site over time. In this study, we propose a deep learning (DL)-based model, Bone Tissue Prediction Generative Adversarial Network (BTP-GAN), which integrates synthetic histological image generation with a biological tissue growth model to simulate bone tissue development over time. Despite a small training dataset, the generated histological images are biologically meaningful and realistic, owing to three algorithmic modules: the Analysis module, Growth module, and Generation module. Moreover, Ordinary Differential Equations (ODE)-Transformer model can learn the temporal patterns and structural characteristics of real tissue images, enabling the generation of histological images according to various specific generative conditions. The BTP-GAN successfully demonstrates the continuous histological change at the same bone tissue site, relying only on a single input histological image. This in-silico animal study may change the perception of the necessity of extensive animal studies in tissue engineering and provide a practical framework for the future advancement of in-silico tissue engineering.