Discovering irregular pupil light responses to chromatic stimuli using waveform shapes of pupillograms

The waveforms of the pupillary light reflex (PLR) can be analyzed in a diagnostic test that allows for differentiation between disorders affecting photoreceptors and disorders affecting retinal ganglion cells, using various signal processing techniques. This procedure has been used on both healthy subjects and patients with age-related macular degeneration (AMD), as a simple diagnostic procedure is required for diagnosis.

The Fourier descriptor technique is used to extract the features of PLR waveform shapes of pupillograms and their amplitudes. To detect those patients affected by AMD using the extracted features, multidimensional scaling (MDS) and clustering techniques were used to emphasize stimuli and subject differences. The detection performance of AMD using the features and the MDS technique shows only a qualitative tendency, however. To evaluate the detection performance quantitatively, a set of combined features was created to evaluate characteristics of the PLR waveform shapes in detail. Classification performance was compared across three categories (AMD patients, aged, and healthy subjects) using the Random Forest method, and weighted values were optimized using variations of the classification error rates. The results show that the error rates for healthy pupils and AMD-affected pupils were low when the value of the coefficient for a combination of PLR amplitudes and features of waveforms was optimized as 1.5. However, the error rates for patients with age-affected eyes was not low.

A classification procedure for AMD patients has been developed using the features of PLR waveform shapes and their amplitudes. The results show that the error rates for healthy PLRs and AMD PLRs were low when the Random Forest method was used to produce the classification. The classification of pupils of patients with age-affected eyes should be carefully considered in order to produce optimum results.