Background: The anterolateral thigh (ALT) flap is widely used for head and neck reconstruction because of its versatility and reliable vascular supply. However, anatomical variability of ALT perforators complicates their consistent identification, which is critical for successful flap harvest. Conventional methods such as Doppler ultrasound often produce false-positive results, making perforator localization challenging. Indocyanine green (ICG) angiography enables real-time intraoperative visualization of vascular flow, but interpretation remains largely subjective. This study integrates artificial intelligence (AI) with ICG angiography to enhance perforator detection, hypothesizing that AI-assisted analysis improves mapping precision and sensitivity.
Methods: This prospective cohort study included 51 patients undergoing ALT flap surgery between February and October 2024. Intraoperative indocyanine green angiography (ICG-A) was performed to identify perforators, followed by grayscale analysis of angiography videos to quantify pixel intensity over time. Perforators were classified as septocutaneous or musculocutaneous and annotated using the Roboflow platform for AI model training. The YOLOv11 object detection algorithm was applied. Model performance was compared with Doppler ultrasound and subjective ICG interpretation in terms of sensitivity and positive predictive value (PPV), with corresponding 95% confidence intervals (CIs). Statistical analysis used the independent t test, with significance set at P < 0.05.
Results: A prototype AI model for ALT perforator detection was developed using ICG-A data. Sensitivity was highest with subjective ICG interpretation (78%; 95% CI, 68%-85%), followed by Doppler ultrasound (53%; 95% CI, 43%-62%) and AI-assisted ICG-A (45%; 95% CI, 26%-65%); PPVs were 28%, 29%, and 21%, respectively. Quantitative pixel analysis showed a mean inflow time of 36 seconds, maximal slope time of 45 seconds, and maximal intensity time of 64 seconds, with a mean maximal intensity of 110 grayscale units. No significant differences were found between musculocutaneous and septocutaneous perforators.
Conclusions: AI-assisted ICG angiography is an emerging tool with potential to support perforator mapping. Although the current AI model demonstrated limited sensitivity, its accuracy can be enhanced by expanding training datasets, integrating temporal fluorescence dynamics, and refining fluorescence-time curve analysis. Future advancements in AI-driven image processing may further optimize intraoperative perforator identification, ultimately improving surgical precision and patient outcomes.
扫码关注我们
求助内容:
应助结果提醒方式:
