OBJECTIVES: This study aimed to build a surgical navigation system based on mixed reality (MR) and optical positioning technique and evaluate its clinical applicability in craniomaxillofacial trauma bone reconstruction. Me-thods We first integrated the software and hardware platforms of the MR-based surgical navigation system and explored the system workflow. The systematic error, target registration error, and osteotomy application error of the system were then analyzed via 3D printed skull model experiment. The feasibility of the MR-based surgical navigation system in craniomaxillofacial trauma bone reconstruction was verified via zygomatico-maxillary complex (ZMC) reduction experiment of the skull model and preliminary clinical study. RESULTS: The system error of this MR-based surgical navigation system was 1.23 mm±0.52 mm, the target registration error was 2.83 mm±1.18 mm, and the osteotomy application error was 3.13 mm±1.66 mm. Virtual surgical planning and the reduction of the ZMC model were successfully conducted. In addition, with the guidance of the MR-based navigation system, the frontal bone defect was successfully reconstructed, and the clinical outcome was satisfactory. CONCLUSIONS The MR-based surgical navigation system has its advantages in virtual reality fusion effect and dynamic navigation stability. It provides a new method for doctor-patient communications, education, preoperative planning, and intraoperative navigation in craniomaxillofacial surgery.
Humans ; Surgical Navigation Systems ; Augmented Reality ; Plastic Surgery Procedures ; Skull/surgery*

Objective: To investigate whether a deep learning-based model using unenhanced computed tomography (CT) at baseline could predict the malignancy of pulmonary nodules. Methods: A deep learning model was trained and applied for the discrimination of pulmonary nodule in Dr. Wise Lung Analyzer. This study retrospectively recruited 130 consecutive participants with pulmonary nodules detected on CT who undergoing biopsy or surgery from May 2009 to June 2017 in Jinling hospital. A total of 136 pulmonary nodules were included in this study, including 86 malignant nodules and 50 benign ones. All patients underwent CT scans 2 times at least, the first scan was defined as baseline and the last scan before the pathological results was defined as final scan. The ROC curve of deep learning model was plotted and the AUCs were calculated. Delong test was used to examine the difference of AUCs baseline and final scan. The nodules were further divided into subsolid nodule group (pure ground-glass nodule and part solid nodule) (n=87) and solid nodule group (n=49). The difference of AUCs at baseline and final scans was evaluated intra two groups. Results: The AUCs of the deep learning model at final and baseline scans were 0.876 and 0.819, respectively. There was no significant difference between them (P=0.075). The result indicated that the model could predict the consequences of pulmonary nodules well at baseline. In small nodules (longest diameter ≤10mm), the AUC at final scan (0.847) was better than it at baseline scan (0.734), but there was no significant difference between them (P=0.058). In solid nodule group, The AUC at final scan (0.932) was better than it at baseline scan (0.835), but there was no significant difference between them (P=0.066). In subsolid nodule group, the deep learning model exhibited consistent performance at final scan (AUC, 0.759) with the baseline scan (AUC, 0.728, P=0.580). Conclusions The deep learning model could predict the malignancy of pulmonary nodules including small ones at baseline, and the model exhibited consistent performance between baseline and final scans in subsolid nodules.