Predicting Malignant Transformation of Choroidal Nevi Using Machine Learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Uveal melanoma (UM) is a rare but aggressive intraocular malignancy.
• Accurate diagnosis of melanocytic choroidal tumors and identification of malignant transformation risk factors are crucial for patient outcomes.
• Current diagnostic methods can be invasive or lack precision.

Approach:

• Developed and trained a machine learning (ML) algorithm using multimodal imaging data, including ultra-widefield fundus imaging and B-scan ultrasonography.
• The ML algorithm assessed key imaging features such as lesion thickness, subretinal fluid, and orange pigment to identify risk factors for UM.
• The algorithm was validated for its ability to classify lesions as UM or choroidal nevi.

Key Points:

• The ML algorithm demonstrated high accuracy in predicting lesion thickness (AUC 0.982) and detecting subretinal fluid (AUC 0.964).
• Sensitivity and specificity for identifying various risk factors ranged from 0.900/0.818 to 1.000/0.727.
• The algorithm successfully differentiated between UM and choroidal nevi based on imaging data.

Conclusions:

• Machine learning can accurately identify UM risk factors from initial multimodal imaging, offering a non-invasive diagnostic approach.
• This ML tool has the potential to prevent unnecessary treatments and reduce metastasis risk.
• The study provides proof of concept for an efficient ML-based system to improve uveal melanoma diagnosis and patient management.