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

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Retinitis pigmentosa (RP) is an inherited retinal disease damaging photoreceptor cells.
• Accurate diagnosis and staging of RP are vital for timely intervention and treatment.
• Current diagnostic methods may lack objectivity in disease staging.

Purpose of the Study:

• To develop and evaluate a machine learning approach for objective diagnosis and staging of retinitis pigmentosa.
• To differentiate between healthy individuals and RP patients across various disease stages.
• To assess the performance of different machine learning algorithms using multifocal electroretinogram (mfERG) data.

Main Methods:

• Amplitude and latency features from mfERG N1, P1, and N2 waves across five rings were extracted.
• Binary and multiclass classification were performed using four machine learning algorithms.
• The Naive Bayes (NB) algorithm was evaluated for its performance in distinguishing RP from healthy controls and staging the disease.

Main Results:

• The Naive Bayes algorithm achieved 99% accuracy in binary classification, distinguishing healthy individuals from each RP stage.
• In multiclass classification, NB achieved 82% accuracy for a four-class problem (including healthy individuals).
• A three-class classification focusing solely on disease stages yielded 76% accuracy with the NB model.

Conclusions:

• The proposed machine learning approach offers objective and accurate staging for retinitis pigmentosa.
• mfERG analysis combined with machine learning can serve as a valuable decision support tool for ophthalmologists.
• This method has the potential to improve clinical practice in managing RP patients.