Abstract
BACKGROUND
Postoperative pain is a common and complex issue that affects patients' recovery and the quality of healthcare. Traditional pain assessment methods-primarily based on self-reporting and clinical observation-are often inadequate, particularly for patients with communication impairments. Deep learning technology offers new opportunities for automatic pain assessment. However, progress in this area is hindered by the limited availability of high-quality clinical datasets and a paucity of studies addressing real-world model deployment. This gap between laboratory research and clinical application requires further study.
METHODS
The study constructed two distinct datasets to capture both clinical and laboratory scenarios. The Clinical Pain Dataset (CPD) includes 3411 facial pain images from 503 postoperative patients, while Simulated Pain Dataset (SPD) contains 1038 images from 51 volunteers. The two datasets were combined to form the Combined Dataset (CD). A pre-trained VGG16 model was used for training and validation. The model's performance on different datasets and pain levels was evaluated using area under the receiver operating characteristic curve (AUROC) and F1 scores.
RESULTS
In the CPD and CD, the model demonstrated its highest performance in identifying severe pain, achieving AUROC values of 0.898 (95 %CI,0.877-0.917) and 0.867 (95 %CI,0.844-0.889), respectively. For overall evaluation, the highest AUROC values were observed in CPD-train (0.898 [95 % CI: 0.877-0.917]) and CD-train (0.917 [95 % CI: 0.883-0.948]) for severe pain classification. Building on these results, a facial pain recognition software was developed based on the model, offering a new option for clinical pain identification.
CONCLUSIONS
The findings indicate that deep learning models leveraging facial expression analysis hold significant potential to recognize varying degrees of pain in clinical settings, especially severe pain. In the future, they could help anesthesiologists monitor postoperative patients' pain levels in real-time, enhancing the quality of medical services.