# Coupling CFD and Machine Learning to Assess Flow Properties in Porous Scaffolds for Tissue Engineering

## TL;DR

This paper uses computational fluid dynamics and machine learning to study how scaffold structure affects fluid flow properties important for tissue engineering.

## Contribution

The novel use of ML to predict flow properties from scaffold topology offers a new design tool for tissue engineering scaffolds.

## Key findings

- ML models achieved R2 above 0.9 for predicting permeability and average wall shear stress from scaffold topology.
- Percentile-based wall shear stress metrics provided insights into environments relevant for bone and cartilage differentiation.
- Pore shape had no significant effect on permeability or average wall shear stress.

## Abstract

Computational fluid dynamics and machine learning (ML) models are employed to investigate the relationships between scaffold topology and key flow parameters, including permeability (k), average wall shear stress (WSSa), and the 25th and 75th percentiles of WSS. Statistical analysis showed that WSSa values are consistent with those found in common scaffold architectures, while percentile-based WSS properties provided insight into shear environments relevant for bone and cartilage differentiation. No significant effect of pore shape was observed on k and WSSa. Correlation analysis revealed that k was positively associated with topological features of the scaffold, whereas WSS metrics were negatively correlated with these properties. ML models trained on six topological and flow inputs achieved a performance of R2 above 0.9 for predicting k and WSSa, demonstrating strong predictive capability based on the topology. Their performance decreased for WSS25% and WSS75%, reflecting the difficulty in capturing more specific shear events. These findings highlight the potential of ML to guide scaffold design by linking topology to flow conditions critical for osteogenesis.