Bayesian Signal Detection of Square-Wave Jerk from Video-Oculographic Recordings

The statistical properties of saccadic intrusions such as square-wave jerks (SWJ) assist with diagnostic assessment of pathological conditions, such as progressive supranuclear palsy. Here, we introduce a novel Bayesian method of detecting square-wave jerks in eye movement traces, and compare it to the current state-of-the-art algorithm1 using a likelihood- based approach. The difference between the two methods consists in their use of information describing non-SWJ saccade pairs. Likelihood-based algorithms rely solely on computing the extent of match between the data and what is theoretically expected from a square-wave saccade pair, where a good (i.e., high likelihood) match is treated as a detection. The Bayesian approach2 takes this a step further, and compares (a) the match between data and a theoretical square-wave saccade pair to (b) the match between data and what is theoretically expected from a non-SWJ saccade pair. The use of additional information contained in (b) allows the algorithm to enhance its discrimination between SWJ and non-SWJ saccade pairs, leading to greater detectability. We test this prediction with noisy simulated eye movement traces.