Machine learning demonstrates clinical utility in distinguishing retinoblastoma from pseudo retinoblastoma with RetCam images.

Abstract

Background: Retinoblastoma is diagnosed and treated without biopsy based solely on appearance (with the indirect ophthalmoscope and imaging). More than 20 benign ophthalmic disorders resemble retinoblastoma and errors in diagnosis continue to be made worldwide. A better noninvasive method for distinguishing retinoblastoma from pseudo retinoblastoma is needed.

Methods: RetCam imaging of retinoblastoma and pseudo retinoblastoma from the largest retinoblastoma center in the U.S. (Memorial Sloan Kettering Cancer Center, New York, NY) were used for this study. We used several neural networks (ResNet-18, ResNet-34, ResNet-50, ResNet-101, ResNet-152, and a Vision Image Transformer, or VIT), using 80% of images for training, 10% for validation, and 10% for testing.

Results: Two thousand eight hundred eighty-two RetCam images from patients with retinoblastoma at diagnosis, 1,970 images from pseudo retinoblastoma at diagnosis, and 804 normal pediatric fundus images were included. The highest sensitivity (98.6%) was obtained with a ResNet-101 model, as were the highest accuracy and F1 scores of 97.3% and 97.7%. The highest specificity (97.0%) and precision (97.0%) was attained with a ResNet-152 model.

Conclusion: Our machine learning algorithm successfully distinguished retinoblastoma from retinoblastoma with high specificity and sensitivity and if implemented worldwide will prevent hundreds of eyes from incorrectly being surgically removed yearly.