Characterisation of nystagmus waveforms in eye-tracker signals

Sammanfattning: This thesis deals with the analysis of eye–tracker signals recorded from nystagmus patients. Nystagmus is an eye movement disorder caused by an underlying condition, and patients who suffer from nystagmus express involuntary oscillating eye move- ments. The oscillatory patterns expressed by these patiens are typically linked to the underlying condition, but it is usually difficult to precisely diagnose each individual. The main focus of this thesis is to develop methods for automatic and robust analysis of nystagmus eye movements. These methods are developed with the purpose of providing diagnostic support for clinicians, or for evaluation of treatment effects.This thesis comprises an introduction and four papers describing various aspects of nystagmus analysis. In all four papers, eye movement signals recorded using an eye tracker are used as input to the proposed methods. In the first paper, a method to robustly calibrate eye–tracker data recorded from nystagmus patients is proposed. Calibration of data from nystagmus patients using video–based systems is difficult since the calibration process relies on an ability to accurately and precisely fixate calibration targets, which is difficult for nystagmus patients. Due to the nystagmus oscillations, it is difficult to obtain calibration results that are acceptable in terms of accuracy. In this work, a novel approach to find outliers in the calibration data is implemented, and a linear Procrustes transformation is used as the calibration mapping function. The results show that the proposed approach leads to reduced gaze estimation variance, and a higher robustness against outliers in the calibration data.In the second paper, a method to model different nystagmus waveform morphologies is presented. This model is used to characterise the nystagmus oscillations and to assert the quality of the analysed eye–tracker signals. The modelling approach is based on a stationary harmonic series, and the signals are modeled in short seg- ments, allowing for tracking of local changes in signal characteristics. Each segment is assessed using a metric referred to as the normalised segment error, which is used to determine whether or not the segment contains measurement disturbances. The results show that the model is well suited to distinguish between nystagmus oscillations and disturbances in the signal.The harmonic model from the second paper is used in the third paper in order to analyse data acquired during both smooth pursuit and fixation eye movements. Smooth pursuit eye movements may carry valuable clinical information, and reliable modelling of smooth pursuit eye movements is therefore of interest. The harmonic model is used to parametrise the different waveforms. Based on the parametrisation, a waveform distance index is defined, which is a metric used to measure similarity between waveforms, as well as for clustering of waveforms. Eleven different clusters are defined using known reference nystagmus waveforms, and all recorded fixation and smooth pursuit waveforms are assigned to one of the eleven cluster centers. The results show that the waveform clustering is robust, is able to distinguish between recordings from different individuals, and is suitable for analysis of smooth pursuit recordings.In the fourth paper, a novel method to combine cycle analysis and morphological classification is proposed. The goal of this work is to provide a diagnostic tool to identify subtle differences between patients, and over time in longer or recurring recordings. The cycle analysis method uses adaptive thresholds in order to detect breaking saccades, fast phases, foveations and slow phases. Eighteen template waveforms are used to create a profile of identified morphologies for each recorded waveform. The method is evaluated against expert annotations from a public dataset. The results show that the method is capable of analysing nystagmus eye movement recordings from both video–based and magnetic scleral search coil techniques. The waveform classification is reliable for both recording techniques.The methods presented in this thesis are used to improve the robustness and reliability for analysis of nystagmus eye movements recorded using an eye–tracker. In total, the four proposed methods constitute a complete framework showing how analysis of nystagmus eye–tracker signals may be used to improve diagnostics in nystagmus patients.