Automatic estimation of hallux valgus angle using deep neural network with axis-based annotation

• DOI: 10.1007/s00256-024-04618-2
• Publication Date: 2024-03-13T08:32:07Z
• Authors: Ryutaro Takeda; Hiroyasu Mizuhara; Akihiro Uchio; Toshiko Iidaka; Kenta Makabe; Taro Kasai; Yasunori Omata; Noriko Yoshimura; Sakae Tanaka; Takumi Matsumoto
• Keywords: Foot (prosody); Proximal phalanx; Mean difference; Mean absolute error

Abstract Objectives We developed the deep neural network (DNN) model to automatically measure hallux valgus angle (HVA) and intermetatarsal (IMA) on foot radiographs. The objective is assess accuracy of by comparing manual measurement ankle surgeons. Materials methods A DNN was predict bone axes first proximal phalanx all metatarsals from fifth in dataset used for development consisted 1798 radiographs collected a population-based cohort patients at our clinic. retrospective validation comprised 92 obtained consecutive visiting mean absolute error (MAE) between automatic measurements median three surgeons compared 3° using one-tailed t -test also inter-rater difference among two-tailed paired -test. Results MAE HVA 1.3° (upper limit 95% CI 1.6°), this significantly smaller than 2.0 ± 0.2° surgeons, demonstrating superior model. In contrast, IMA 0.8° 1.0°) that showed no significant 1.0 0.1° Conclusion Our demonstrated ability with an comparable specialists.