Currently, orthotic design relies heavily on surgeons' experience, lacking consistency. we integrated AI, healthcare and interaction design to develop a neural network model for predicting corrective forces in scoliosis orthoses. A database comprising Cobb angle, flexibility, BMI, and corrective force data from 260 patients was constructed. A backpropagation neural network (BP) was constructed, and subsequently integrated with genetic algorithms (GA), particle swarm optimization (PSO), and gradient descent (GD) to establish four distinct models: BP, GA-BP, PSO-BP, and GD-BP. Based on metrics including the fit and error of the training and test sets, as well as the residuals between actual and predicted values, a systematic comparison and analysis were conducted on the four models. The GD-BP model demonstrated the optimal prediction performance. Finally, a software tool for predicting corrective forces was developed based on this model. A field trial involving 28 patients showed that the fit between the model's predicted values and actual values is 0.85, thereby confirming the effectiveness of the model. The model can calculate appropriate correction force parameters based on patient physical data, making correction force settings more scientific. Additionally, it integrates artificial intelligence with medical expertise to establish a precise and efficient methodology for designing scoliosis orthoses.
Li et al. (Wed,) studied this question.