Computer-aided classification of sickle cell retinopathy using quantitative features in optical coherence tomography...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Sickle cell retinopathy (SCR) diagnosis lacks standardized quantitative optical coherence tomography angiography (OCTA) interpretation.
• Developing objective classification methods for SCR is crucial for patient management.

Purpose of the Study:

• To demonstrate computer-aided classification of SCR using quantitative OCTA features.
• To evaluate the performance of different machine learning classifiers for SCR detection and staging.

Main Methods:

• Quantitative OCTA features extracted: blood vessel tortuosity (BVT), blood vessel diameter (BVD), vessel perimeter index (VPI), foveal avascular zone (FAZ) area, FAZ contour irregularity, and parafoveal avascular density (PAD).
• Machine learning classifiers evaluated: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Discriminant Analysis.
• Performance metrics: Sensitivity, specificity, and accuracy for SCR vs. control and mild vs. severe SCR classification.

Main Results:

• Combined OCTA features demonstrated superior classification performance compared to individual features.
• All classifiers achieved high accuracy (average 95%) for SCR vs. control classification.
• SVM achieved the highest accuracy (97%) for mild vs. severe SCR classification, outperforming KNN (95%) and Discriminant Analysis (88%).

Conclusions:

• Quantitative OCTA features combined with machine learning offer a robust approach for SCR classification.
• Computer-aided OCTA analysis shows promise for objective and accurate diagnosis and staging of SCR.
• This methodology can potentially aid in early detection and management of sickle cell retinopathy.