Computed Tomography Angiography and B-Mode Ultrasonography under Artificial Intelligence Plaque Segmentation Algorithm in the Perforator Localization for Preparation of Free Anterolateral Femoral Flap.
CONTRAST MEDIA & MOLECULAR IMAGING 2022;
2022:4764177. [PMID:
36247860 PMCID:
PMC9534661 DOI:
10.1155/2022/4764177]
[Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 05/25/2022] [Revised: 07/15/2022] [Accepted: 08/05/2022] [Indexed: 01/26/2023]
Abstract
This research was aimed to investigate the accuracy of U-shaped network (UNet)-based computed tomography angiography (CTA) and B-mode ultrasonography (US) in the perforator localization of free anterolateral thigh flap (ALTF). Based on UNet, a fusion of deep supervision mechanism, squeeze-and-excitation module, and attention mechanism was introduced to optimize the algorithm. Then, a CTA segmentation model, DA-UNet, was established. The segmentation performance of DA-UNet and other algorithms was compared under the same conditions. 30 patients who were planned to receive ALTF surgery were selected as the research objects. According to different preoperative localization methods, they were divided into group A (CTA) and group B (B-mode US), 15 cases in each group. Combined with the actual situation during surgery, the diagnostic accordance rate, sensitivity (Sen), specificity, and the distance between the perforator location and the actual location were compared between the two groups. The Dice coefficient, Jaccard index, Sen, the area under curve (AUC), and average Hausdorff distance (AVD) of the DA-UNet segmentation algorithm were 80.70%, 69.97%, 77.56%, 0.887, and 2.48, respectively. These results were significantly better than those of other algorithms (P < 0.05). In group A, the diagnostic accordance rate, Sen, and specificity of patients were 96.55%, 90.52%, and 73.58%, respectively, which were higher than 91.53%, 81.36%, and 15.60% of patients in group B significantly (P < 0.05). There was no statistical difference in the distance between the perforator location and the actual location (P > 0.05). It showed that the accuracy of CTA under the UNet-based DA-UNet segmentation model in the perforator localization of ALTF was better than that of B-mode US. Thus, a reference could be provided for the preparation of free ALTF and its clinical application.
Collapse