Abstract Postoperative flap monitoring is critical for early detection and salvage of compromised flaps. As shorter hospital stays and outpatient procedures become common, accurate and user-friendly tools for remote flap viability assessment are needed. This study describes the development and validation of FlapCheck.ai, an automated artificial intelligence (AI) custom vision model that classifies flap viability from nonstandardized clinical images. Images of postoperative flaps were obtained through a single-center retrospective review and literature search and labeled as healthy or compromised. Two hundred nine images were split into training (80%) and testing (20%) sets with class balance preserved. Data augmentation (rotation, flipping, brightness, contrast adjustments) expanded the training set to 1,432 training images. The model was trained using Microsoft Azure Custom Vision and evaluated on the testing dataset. Performance metrics included accuracy, sensitivity, specificity, precision, F1 score, and area under the receiver operating characteristic curve (AUC-ROC) with 95% confidence intervals (CI). The model correctly classified 40 of 41 test images (30/31 healthy, 10/10 compromised), yielding an accuracy of 97.6% (95% CI: 87.4–99.6), sensitivity of 100% (95% CI: 72.2–100), specificity of 96.8% (95% CI: 83.8–99.4), precision of 90.9% (95% CI: 70.0–100), F1 score of 95.2% (95% CI: 82.4–100), and an AUC of 0.997. A fully automated AI model was successfully developed and validated for postoperative flap viability assessment using nonstandardized images. FlapCheck.ai demonstrated excellent diagnostic performance and may enable reliable outpatient flap monitoring and earlier detection of flap compromise. Future work will expand datasets and evaluate the impact in prospective studies.
Karir et al. (2026) studied this question.