Estimation of the Radiographic Parameters for Hallux Valgus from Photography of the Feet Using a Deep Convolutional Neural Network

Abstract

Hallux valgus (HV) is one of the most common forefoot deformities. Early diagnosis and proper evaluation of HV are important for timely management of HV. The purpose of our study was to estimate the radiographic parameters of HV using deep learning to assess the agreement of the predicted measurement with the actual radiographic measurement. There were 131 patients enrolled in this study. A total of 248 radiographs and 337 photographs of the feet were acquired. Radiographic parameters including the HV angle (HVA), M1-M2 angle, and M1-M5 angle were measured. We constructed a convolutional neural network using Xception and made it a regression model. Then, we fine-tuned the model using images of the feet and the radiographic parameters. The coefficient of determination (R2) and root mean squared error (RMSE) were calculated to evaluate the performance of the model. The radiographic parameters, including the HVA, M1-M2 angle, and M1-M5 angle were predicted with an R2=0.684, RMSE=7.91; R2=0.573, RMSE=3.29; R2=0.381, RMSE=5.80, respectively. The present study demonstrated that our model was able to predict the radiographic parameters of HV from photography. This study shows a potential application of deep learning for HV screening.

Validity and reliability of hallux valgus angle measurement on smartphone digital photographs

“…Recently, Inoue et al [10] attempted to estimate the radiographic parameters for HV, namely HVA, IMA 1-2, and IMA 4-5, from photography using a deep convolutional neural network (CNN). They produced a CNN model and the estimated HVA with their automatic prediction model without user intervention.…”