Monitoring Preoperative Surgical Hand Washing Of Ophthalmic Medical Staff Based On Artificial Intelligence Behavior Recognition Technology
Published 2025 - 43rd Congress of the ESCRS
Reference: PO272 | Type: Free paper | DOI: 10.82333/ytrk-zv15
Authors: Sofia Bielsa Alonso* 1 , Maite Sisquella 1 , Laura Gonzalez 1 , Ana Nolla 1 , Sandra Suescun 1 , Mar García 1 , Ruth Sintes 1 , Alvaro Terroba 1 , Jose Luis Güell 1
1IMO Barcelona,Barcelona,Spain
Purpose
Constructing a hand washing monitoring system based on artificial intelligence behavior recognition technology to evaluate the accuracy of preoperative surgical hand washing action recognition for ophthalmic medical staff
Setting
Cohort study
Methods
Data Processing : 2200 preop handwashing videos (200 standardized) analyzed for 25 action categories/5 item categories. 2000 videos split into 8:1:1 sets (train/validation/test) for deep learning model development. Model Evaluation : Performance metrics: precision, recall, F1-score, mAP (IoU thresholds). Personnel classification: single/multi-person scenarios analyzed for prediction accuracy.
Results
Model Performance : Single-Scenario : mAP=0.9810 (IoU=0.5), F1=0.9800, Accuracy=1.0000, Recall=0.9900. Multi-Scenario : mAP=0.8699 (IoU=0.5), F1=0.8760, Accuracy=0.8937, Recall=0.8699. Key Findings : Wrist movements/arm scrubbing/disinfectant use show 2%-3% misjudgment rate. Multi-scenario errors: motion occlusion (10%), blur (10%), background (10%). Handwashing standardization accuracy: Test set=93.43%.
Conclusions
The hand washing monitoring system based on artificial intelligence behavior recognition technology can achieve perfect detection of most hand washing actions in single person scenes. However, in multi person scenes, complex backgrounds, and motion occlusion situations, misjudgment may occur, and further optimization of data and models is needed to improve prediction accuracy.