Leveraging Deep Learning Models for Retinoblastoma Detection

Retinoblastoma, a rare pediatric eye cancer, requires early detection to prevent vision loss and improve survival rates. This study evaluates the performance of five deep learning models—MobileNetV2, EfficientNetB0, ResNet50, DenseNet121, and VGG16—for retinoblastoma detection using retinal images. After preprocessing and data augmentation, models were assessed based on accuracy, precision, recall, F1-score, and area under the Receiver Operating Characteristic (ROC) curve. The results show that EfficientNetB0 (99.0% accuracy, 100% recall) and DenseNet121 (99.7% accuracy, 100% recall) outperform other models, making them ideal for clinical applications. MobileNetV2 and ResNet50 provide a balance between computational efficiency and performance, while VGG16 struggles with generalization. This study highlights the potential of deep learning in improving automated retinoblastoma diagnosis.