Denoising eye movement signals using a novel signal processing framework to preserve oculomotor integrity in dyslexia.

Eye movement analysis serves as a powerful tool for early dyslexia detection. Signal quality issues, signal degradation from noise, drift, and motion artifacts, remain a major obstacle to reliable eye movement analysis. To overcome these challenges and enhance the quality of eye movement data, we proposed a two-stage framework consisting of primary signal denoising and Adaptive recursive enhancement. In the first stage, denoising using filters such as the Savitzky-Golay (SG) filter, the Infinite Impulse Response (IIR) filter, the finite impulse response (FIR) filter, and the Median filter are applied, with the output exhibiting the highest signal-to-noise ratio (SNR) selected for further analysis. If initial filtering is unable to improve SNR enough, the pipeline moves on to the second stage, where Kalman filters with the Constant Velocity Model, Bayesian Model, and Constant Acceleration Model refine the signal while preserving features like saccades. Quantitative evaluation of the pipeline using metrics, the IIR filter had the highest SNR in the first stage, while CAMKF had the highest SNR in the second stage. Compared to the signal filter method, the two-stage approach consistently yielded higher SNR and better preservation of eye movement features, improving signal quality and enabling more precise dyslexia related analysis.