Photo Credit: flik47

The following is a summary of “Deep learning-based system for automatic identification of benign and malignant eyelid tumours,” published in the May 2025 issue of British Journal of Ophthalmology by Fan et al. 

Researchers conducted a retrospective study to create a deep learning-based system for automatically specifying and categorizing benign and malignant eyelid tumors to improve diagnostic accuracy and efficiency.   

They utilized a dataset containing photographs of normal eyelids and benign and malignant eyelid tumors. The dataset was unsystematically divided into training and validation sets in an 8:2 ratio. The training set was used to train 8 convolutional neural network models to classify normal, benign, and malignant eyelid tumors, including VGG16, ResNet50, Inception-v4, EfficientNet-V2-M, and their variants. The validation set was used to evaluate and compare the performance of each deep learning model.   

The results showed that all 8 convolutional neural network models reached an average accuracy above 0.746 in classifying normal eyelids, benign, and malignant eyelid tumors. Sensitivity and specificity averages exceeded 0.790 and 0.866, respectively. The mean area under the receiver operating characteristic curve (AUC) across all models was greater than 0.904. The dual-path Inception-v4 model achieved the best results, with an AUC of 0.930 (95% CI 0.900 to 0.954) and an F1-score of 0.838 (95% CI 0.787 to 0.882).   

Investigators concluded that a deep learning-based system demonstrated considerable promise for enhancing eyelid tumor diagnosis as a dependable and efficient tool for clinical application. 

Source: bjo.bmj.com/content/early/2025/05/10/bjo-2025-327127