Early detection and staging of retinitis pigmentosa using multifocal electroretinogram parameters and machine learning algorithms

Retinitis pigmentosa is an inherited retinal disease caused by damage to photoreceptor cells. Diagnosis and staging of this disease are crucial for early intervention and effective treatment planning. In this study, the amplitude and latency features of N1, P1, and N2 waves obtained from multifocal electroretinogram responses over five rings were used with binary and multiclass classification methods using four different machine learning algorithms to distinguish retinitis pigmentosa patients from healthy individuals and to evaluate the stages of the disease. Binary classifications were performed for six different groups, and the Naive Bayes (NB) algorithm performed the best on all evaluation metrics, achieving 99% accuracy in distinguishing healthy individuals from each disease stage. Furthermore, multiclass classification was applied in two different steps. In the first step, the Naive Bayes model achieved 82% accuracy in four-class classification, including healthy individuals. Considering the near-perfect separability of healthy individuals, in the second step, a three-class classification including only disease stages was performed, and the model achieved 76% accuracy. These results indicate that the proposed approach provides objective and accurate staging for retinitis pigmentosa and can serve as a valuable decision support system to assist ophthalmologists in clinical practice.