Faculty Bibliography

Alzheimer's disease (AD) has been considered one of the most challenging forms of dementia. The earlier the people are diagnosed with AD, the easier it is for doctors to find a treatment. Based on the previous literature summarizing the research results on the relationship between eye movement and AD before 2013, this paper reviewed 34 original eye movements research papers only closely related to AD published in the past ten years and pointed out that the prosaccade (4 papers) and antisaccade (5 papers) tasks, reading tasks (3 papers), visual search tasks (3 papers) are still the research objects of many researchers, Some researchers have looked at King-Devick tasks (2 papers), reading tasks (3 papers) and special tasks (8 papers), and began to use combinations of different saccade tasks to detect the relationship between eye movement and AD, which had not been done before. These reflect the diversity of eye movement tasks and the complexity and difficulty of the relationship between eye movement and AD. On this basis, the current processing and analysis methods of eye movement datasets are analyzed and discussed in detail, and we note that certain key data that may be especially important for the early diagnosis of AD by using eye movement studies cannot be miss-classified as noise and removed. Finally, we note that the development of methods that can accurately denoise and classify and quickly process massive eye movement data is quite significant for detecting eye movements in early diagnosis of AD.

Saccadic slowing as a component of supranuclear saccadic gaze palsy is an important diagnostic sign in multiple neurologic conditions, including degenerative, inflammatory, genetic, or ischemic lesions affecting brainstem structures responsible for saccadic generation. Little attention has been given to the accuracy with which clinicians correctly identify saccadic slowing. We compared clinician (n = 19) judgements of horizontal and vertical saccade speed on video recordings of saccades (from 9 patients with slow saccades, 3 healthy controls) to objective saccade peak velocity measurements from infrared oculographic recordings. Clinician groups included neurology residents, general neurologists, and fellowship-trained neuro-ophthalmologists. Saccades with normal peak velocities on infrared recordings were correctly identified as normal in 57% (91/171; 171 = 9 videos × 19 clinicians) of clinician decisions; saccades determined to be slow on infrared recordings were correctly identified as slow in 84% (224/266; 266 = 14 videos × 19 clinicians) of clinician decisions. Vertical saccades were correctly identified as slow more often than horizontal saccades (94% versus 74% of decisions). No significant differences were identified between clinician training levels. Reliable differentiation between normal and slow saccades is clinically challenging; clinical performance is most accurate for detection of vertical saccade slowing. Quantitative analysis of saccade peak velocities enhances accurate detection and is likely to be especially useful for detection of mild saccadic slowing.

Convolutional neural networks (CNNs) provide impressive empirical success in various tasks; however, their inner workings generally lack interpretability. In this paper, we interpret shallow CNNs that we have trained for the task of positive sparse signal denoising. We identify and analyze common structures among the trained CNNs. We show that the learned CNN denoisers can be interpreted as a nonlinear locally-adaptive thresholding procedure, which is an empirical approximation of the minimum mean square error estimator. Based on our interpretation, we train constrained CNN denoisers and demonstrate no loss in performance despite having fewer trainable parameters. The interpreted CNN denoiser is an instance of a multivariate spline regression model, and a generalization of classical proximal thresholding operators.

This paper proposes a Kalman filter framework for signal denoising that simultaneously utilizes conventional linear time-invariant (LTI) filtering and total variation (TV) denoising. In this approach, the desired signal is considered to be a mixture of two distinct components: a band-limited (e.g., low-frequency component, high-frequency component) signal and a sparse-derivative signal. An iterative Kalman filter/smoother approach is formulated where zero-phase LTI filtering is used to estimate the band-limited signal and TV denoising is used to estimate the sparse-derivative signal.

(1) Background: The King-Devick (KD) rapid number naming test is sensitive for concussion diagnosis, with increased test time from baseline as the outcome measure. Eye tracking during KD performance in concussed individuals shows an association between inter-saccadic interval (ISI) (the time between saccades) prolongation and prolonged testing time. This pilot study retrospectively assesses the relation between ISI prolongation during KD testing and cognitive performance in persistently-symptomatic individuals post-concussion. (2) Results: Fourteen participants (median age 34 years; 6 women) with prior neuropsychological assessment and KD testing with eye tracking were included. KD test times (72.6 ± 20.7 s) and median ISI (379.1 ± 199.1 msec) were prolonged compared to published normative values. Greater ISI prolongation was associated with lower scores for processing speed (WAIS-IV Coding, r = 0.72, p = 0.0017), attention/working memory (Trails Making A, r = -0.65, p = 0.006) (Digit Span Forward, r = 0.57, p = -0.017) (Digit Span Backward, r= -0.55, p = 0.021) (Digit Span Total, r = -0.74, p = 0.001), and executive function (Stroop Color Word Interference, r = -0.8, p = 0.0003). (3) Conclusions: This pilot study provides preliminary evidence suggesting that cognitive dysfunction may be associated with prolonged ISI and KD test times in concussion.

Interictal epileptiform discharges (IEDs) can impair memory. The properties of IEDs most detrimental to memory, however, are undefined. We studied the impact of temporal and spatial characteristics of IEDs on list learning. Subjects completed a memory task during intracranial EEG recordings including hippocampal depth and temporal neocortical subdural electrodes. Subjects viewed a series of objects, and after a distracting task, recalled the objects from the list. The impacts of IED presence, duration, and propagation to neocortex during encoding of individual stimuli were assessed. The effects of IED total number and duration during maintenance and recall periods on delayed recall performance were also determined. The influence of IEDs during recall was further investigated by comparing the likelihood of IEDs preceding correctly recalled items vs. periods of no verbal response. Across 6 subjects, we analyzed 28 hippocampal and 139 lateral temporal contacts. Recall performance was poor, with a median of 17.2% correct responses (range 10.4-21.9%). Interictal epileptiform discharges during encoding, maintenance, and recall did not significantly impact task performance, and there was no significant difference between the likelihood of IEDs during correct recall vs. periods of no response. No significant effects of discharge duration during encoding, maintenance, or recall were observed. Interictal epileptiform discharges with spread to lateral temporal cortex during encoding did not adversely impact recall. A post hoc analysis refining model assumptions indicated a negative impact of IED count during the maintenance period, but otherwise confirmed the above results. Our findings suggest no major effect of hippocampal IEDs on list learning, but study limitations, such as baseline hippocampal dysfunction, should be considered. The impact of IEDs during the maintenance period may be a focus of future research.

The quantitative analysis of saccades in eye movement data unveils information associated with intention, cognition, and health status. Abnormally slow saccades are indicative of neurological disorders and often imply a specific pathological disturbance. However, conventional saccade detection algorithms are not designed to detect slow saccades, and are correspondingly unreliable when saccades are unusually slow. In this article, we propose an algorithm that is effective for the detection of both normal and slow saccades. The proposed algorithm is partly based on modeling saccadic waveforms as piecewise-quadratic signals. The algorithm first decreases noise in acquired eye-tracking data using optimization to minimize a prescribed objective function, then uses velocity thresholding to detect saccades. Using both simulated saccades and real saccades generated by healthy subjects and patients, we evaluate the performance of the proposed algorithm and 10 other detection algorithms. We show the proposed algorithm is more accurate in detecting both normal and slow saccades than other algorithms.