Classification of Retinitis Pigmentosa Stages Based on Machine Learning by Fusion of Image Features of VF and MfERG Maps

Retinitis Pigmentosa (RP) is a progressive retinal disorder that leads to vision loss and blindness. Accurate staging of RP is crucial for effective treatment planning and disease management. This study aims to develop an objective and reliable method for RP staging by integrating handcrafted features extracted from visual field (VF) grayscale and multifocal electroretinography (mfERG) P1 wave amplitude maps using machine-learning models. Four machine-learning models were evaluated using features derived from VF grayscale maps (GLCM and gray tone features) and mfERG P1 amplitude maps (RGB and HSV features). Additionally, feature selection was performed using the Random Forest (RF) algorithm to identify the most relevant features. The experimental results showed that the Support Vector Machine (SVM) model achieved the highest classification performance with 98.39% accuracy, 98.26% precision, 98.55% recall, 98.41% F1 score, and 99.17% specificity using the seven most important features: RGB Entropy_R, GLCM Contrast_90°, RGB Std_R, GLCM Homogeneity_90°, RGB Energy_R, Histogram Kurtosis, and GLCM Energy_90°. These findings demonstrate that fusing grayscale and amplitude maps provides an effective approach for RP staging. The proposed method may serve as an objective, automated decision-support tool for ophthalmologists, enhancing clinical evaluations and enabling personalized treatment strategies for RP patients.