Comprehensive Summary
This study by Takeda et al. tests their AI smartphone application, which measures the hallux valgus angle (HVA) and intermetatarsal angles from radiographic images on a monitor. Three different users (a certified orthopedic surgeon, a resident and a nonhealthcare professional) tested the application with 92 radiographs. The results from the application were compared with the clinical measurements done by experienced foot & ankle surgeons. The final mean absolute errors of the HVA measured by the three users were 1.1°, 1.3°, and 1.4°. Moreover, the intermetatarsal angle measurements displayed similar accuracy. These values are significantly smaller than the threshold error of 3°. The researchers remarked that the difference in clinical experience did not significantly influence the accuracy of the measurements, implying that anyone could use the tool. As a result, the researchers concluded that the application is satisfying enough to be used in clinical settings and research.
Outcomes and Implications
Previously, the researchers developed the application as an alternative to a neural network model they developed. Their model, however, was not practical as it required Digital Imaging and Communications in Medicine data. As a result, they developed a neural network application which only requires image input from a camera. This tool can therefore be given to users to be used in a clinical setting or research, regardless of their clinical expertise. Nevertheless, the researchers acknowledge that their program cannot yet be used to analyze postoperative radiograph, needing another neural network to be trained for that purpose.