Artificial intelligence-based numerical analysis of the quality of facial reanimation: A comparative retrospective cohort study between one-stage dual innervation and single innervation

In recent years, there have been many advances in the functional reconstruction of facial palsy patients. Nerve transfers are indicated in short-standing facial paralysis with loss of the proximal facial nerve signal. Cross face nerve graft (CFNG), hypoglossal nerve, and masseter nerve are commonly transferred neural sources (Biglioli et al., 2012b, 2018). Meanwhile, free muscle transfer is a good modality for long-standing facial palsy patients (Oh et al., 2019; Boonipat et al., 2020b; Singh Kalra et al., 2022). Selection of functional muscle and selection of donor nerve are the main concerns of plastic surgeons. Functional muscles include the Latissimus dorsi muscle, gracilis muscle, temporalis muscle, and rectus abdominis (Okazaki et al., 2015; Park et al., 2022), which have recently been standardized as functional gracilis muscle transfer (FGMT) (Boonipat et al., 2020b). Although the aesthetic and functional outcomes of the surgery are improved, the aesthetic results, including facial symmetry, are still unpredictable after free muscle transfer.

Various donor nerve options and surgical options are available. Among them, CFNG has been traditionally used as a method using the contralateral facial nerve, but has the disadvantage of high donor site morbidity and requiring two surgeries or long recovery time. Although the use of the masseteric nerve as a donor nerve was initially introduced by Escat and Viela in 1925 (Escat and Viela, 1925), no clinical series have reported on its efficacy to date. However, Biglioli et al. have presented clinical series and advantages of this technique over the traditionally used CFNG, leading to its widespread use (Biglioli et al., 2012b). This approach has proved effective in overcoming the drawbacks associated with the use of the contralateral facial nerve, and its large neural input provides additional benefits. Despite these advantages, the optimal technique for facial nerve reconstruction remains uncertain, as dual innervation using both CFNG and the masseteric nerve has been proposed. To accurately compare the effectiveness of these techniques, a reliable evaluation method must be established. The evaluation method should be quantitative and objective. The evaluation of facial paralysis and post-treatment evaluation have also been done in various ways, and the use of software has been the most commonly used method. Representative software includes Facegram and Emotrics (Horta et al., 2014; Dong et al., 2018; Oh et al., 2019; Kim et al., 2020; Krag et al., 2021). However, in the case of Emotrics, a software using an automated facial landmark tracking algorithm through machine learning, relatively few facial landmarks were characterized and evaluated, and landmark selection was based on 2 dimensions (Greene et al., 2019; Dusseldorp and Hadlock, 2022). Recently, studies have been developed further and facial expressions measured by applying artificial intelligence (AI) (Boonipat et al., 2020a). However, this study aimed to determine emotion in facial expressions before and after facial reanimation surgery through AI that reads emotion from facial expressions.

In this study, an evaluation method is proposed that uses deep learning to obtain 468 three-dimensional (3D) facial landmarks from a patient's clinical photos, which are two-dimensional (2D) images. The coordinates of the landmarks that are of interest are then used to determine the degree of symmetry of the face(Kim et al., 2022). The symmetry score was calculated using this method. As a more quantitative way to quantify facial paralysis, the degree of improvement of facial paralysis before and after surgery was compared.

The study compared postoperative improvement in single and dual innervation groups and proposes a new method of quantifying facial symmetry using 3D landmarks extracted through deep learning.

留言 (0)

沒有登入
gif