Development of an Automatic Segmentation System for Anterolateral Thigh Flap Perforators in Maxillofacial Reconstruction

Abstract

Introduction : The anterolateral thigh (ALT) flap is one of the most preferred flaps for maxillofacial reconstruction and for finding a skin perforator in an important first step in flap design. Computed tomography angiography (CTA) can be used through image mapping, contributing to the establishment of an appropriate flap design plan and reducing surgical time. However, preoperative mapping is time-consuming and can be affected by the image quality and clinician’s skill. This study aimed to develop an automatic segmentation system for ALT flap perforators using a convolutional neural network in CTA and evaluate its efficiency and accuracy. Material & Methods : This study enrolled 80 patients who underwent CTA for maxillofacial reconstruction. A single researcher performed mapping of bilateral lateral femoral circumflex artery perforators on uploaded CTA using AVIEW Modeler (version 1.1.42.7, Coreline Software, Seoul, Korea). Training for the development of automatic segmentation system was conducted using DeepLabv2 and 3D ResNet152. Dice Similarity Coefficient (DSC) and Jaccard Similarity Coefficient (JSC) values were calculated, and the distance difference between two points was measured. Furthermore, quadrant distribution analysis was conducted. Results : As a result of the training, the DSC and JSC values were 69.67 ± 1.48 and 67.81 ± 1.70, respectively. The test results for the 13 dataset were DSC with 69.34 ± 0.83 and JSC with 67.47 ± 0.96. The distance difference between manual and automatic detection was 38.28 ± 15.52 mm on the left side and 31.96 ± 18.11 mm on the right side. Quadrant distribution was most prevalent in the distosuperior area. Conclusion : The reasonable results indicate that the automatic segmentation system for ALT flap perforator has potential for clinical application.