Deep learning-based system for automatic identification of benign and malignant eyelid tumours

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Eyelid tumors require accurate and efficient diagnosis.
• Current diagnostic methods can be time-consuming and may lack precision.

Purpose of the Study:

• To develop a deep learning system for automated identification and classification of benign and malignant eyelid tumors.
• To enhance diagnostic accuracy and efficiency for eyelid tumors.

Main Methods:

• Utilized a dataset of normal eyelids and eyelid tumors, randomly split into training (80%) and validation (20%) sets.
• Trained eight convolutional neural network (CNN) models, including VGG16, ResNet50, Inception-v4, and EfficientNet-V2-M variants.
• Evaluated and compared model performance using the validation dataset.

Main Results:

• All eight models achieved average accuracy >0.746, sensitivity >0.790, and specificity >0.866.
• The mean area under the receiver operating characteristic curve (AUC) for all models exceeded 0.904.
• The dual-path Inception-v4 network showed the highest performance with an AUC of 0.930 and an F1-score of 0.838.

Conclusions:

• The developed deep learning system demonstrates significant potential for improving eyelid tumor diagnosis.
• This system offers a reliable and efficient tool for clinical application.
• Future research will focus on validating the model with larger datasets and integrating it into clinical workflows.