Unsupervised learning of stationary and switching dynamical system models from Poisson observations.

Objective. Investigating neural population dynamics underlying behavior requires learning accurate models of the recorded spiking activity, which can be modeled with a Poisson observation distribution. Switching dynamical system models can offer both explanatory power and interpretability by piecing together successive regimes of simpler dynamics to capture more complex ones. However, in many cases, reliable regime labels are not available, thus demanding accurate unsupervised learning methods for Poisson observations. Existing learning methods, however, rely on inference of latent states in neural activity using the Laplace approximation, which may not capture the broader properties of densities and may lead to inaccurate learning. Thus, there is a need for new inference methods that can enable accurate model learning.Approach. To achieve accurate model learning, we derive a novel inference method based on deterministic sampling for Poisson observations called the Poisson Cubature Filter (PCF) and embed it in an unsupervised learning framework. This method takes a minimum mean squared error approach to estimation. Terms that are difficult to find analytically for Poisson observations are approximated in a novel way with deterministic sampling based on numerical integration and cubature rules.Main results. PCF enabled accurate unsupervised learning in both stationary and switching dynamical systems and largely outperformed prior Laplace approximation-based learning methods in both simulations and motor cortical spiking data recorded during a reaching task. These improvements were larger for smaller data sizes, showing that PCF-based learning was more data efficient and enabled more reliable regime identification. In experimental data and unsupervised with respect to behavior, PCF-based learning uncovered interpretable behavior-relevant regimes unlike prior learning methods.Significance. The developed unsupervised learning methods for switching dynamical systems can accurately uncover latent regimes and states in population spiking activity, with important applications in both basic neuroscience and neurotechnology.

Neurofilament Light Chain in Aqueous Humor as a Marker of Neurodegeneration in Glaucoma.

Purpose: Neurofilament light chain (NfL) is a neuronal cytoskeletal protein that has been identified as a marker of neurodegeneration in diseases of the central nervous system. In this study, we investigated whether NfL in the aqueous humor (AH) can serve as a marker of neurodegeneration in glaucoma in a racially diverse North American population. Design: Single-center, case-control study. Participants: We enrolled patients with various types and stages of glaucoma undergoing planned ophthalmic surgery as part of their routine care and compared them with patients without glaucoma undergoing phacoemulsification for age-related cataract. Methods: We collected AH from 39 glaucoma patients and 10 patients without glaucoma. AH NfL was quantified using the Single-Molecule Array (Simoa)® NF-light assay (Quanterix). Demographic information, such as age, body mass index, sex, and self-reported race, as well as clinical information, such as pre-operative intraocular pressure (IOP), maximum IOP, and number of pre-operative glaucoma medications, was obtained by reviewing the medical record. Main Outcome Measures: Levels of AH NfL. Results: In a model controlling for age and body mass index (BMI), NfL was significantly elevated in AH from glaucoma patients (mean: 429 pg/mL; standard deviation [SD]: 1136 pg/mL) compared to AH from patients without glaucoma (mean: 3.1 pg/mL; SD: 1.9 pg/mg): P = 0.002. Higher AH NfL was associated with higher maximum IOP (R = 0.44, P = 0.005), higher pre-operative IOP (R = 0.46, P = 0.003), and more pre-operative glaucoma medications (Rs = 0.61, P < 0.001). There was no association between AH NfL and Humphrey visual field mean deviation (R = -0.20, P = 0.220), retinal nerve fiber layer thickness as measured with optical coherence tomography (R = 0.07, P = 0.694), or glaucoma stage (Rs = 0.015, P = 0.935). Conclusion: Our findings suggest that AH NfL may have clinical utility as a marker of glaucomatous neurodegeneration.

Neurodegeneration Markers Galectin-3 and Apolipoprotein E Are Elevated in the Aqueous Humor of Eyes With Glaucoma.

Purpose: Galectin-3 (Gal-3) and apolipoprotein E (APOE) are markers of activated microglia in neurodegenerative diseases of the central nervous system, whose targeting is protective in mouse models of glaucoma. In this study, we examined levels of Gal-3 and APOE in human aqueous humor (AH) and defined their clinical associations with glaucoma. Methods: We collected AH from 59 glaucoma patients and 15 controls at the start of planned ophthalmic surgery. Gal-3 and APOE levels were quantified by enzyme-linked immunosorbent assay. Total protein in AH was quantified by bicinchoninic acid assay. Significant associations between Gal-3, APOE, and clinical covariates were defined using univariate and multivariate linear regression models. Results: Gal-3 and APOE levels were significantly elevated in the AH of glaucoma patients compared to controls (P = 0.004 and P < 0.001, respectively). Gal-3 and APOE were positively correlated across the entire cohort (r = 0.65, P = 6.2E-9). No association was observed between Gal-3 and total protein or APOE and total protein (P = 0.35 and P = 0.50, respectively), indicating that their levels were not increased in glaucomatous AH due to nonspecific protein accumulation. Multivariate linear regression modeling revealed significant associations between Gal-3 and maximum recorded intraocular pressure (P = 0.009) and between APOE and number of past ophthalmic surgeries (P = 0.031). Conclusions: We demonstrate that Gal-3 and APOE are significantly elevated in the AH of eyes with glaucoma and are associated with a history of poorly controlled disease. Translational Relevance: Gal-3 and APOE in AH may inform clinical decision-making as quantifiable readouts of microglial activation in eyes with glaucoma.

Modeling and inference methods for switching regime-dependent dynamical systems with multiscale neural observations.

Objective.Realizing neurotechnologies that enable long-term neural recordings across multiple spatial-temporal scales during naturalistic behaviors requires new modeling and inference methods that can simultaneously address two challenges. First, the methods should aggregate information across all activity scales from multiple recording sources such as spiking and field potentials. Second, the methods should detect changes in the regimes of behavior and/or neural dynamics during naturalistic scenarios and long-term recordings. Prior regime detection methods are developed for a single scale of activity rather than multiscale activity, and prior multiscale methods have not considered regime switching and are for stationary cases.Approach.Here, we address both challenges by developing a switching multiscale dynamical system model and the associated filtering and smoothing methods. This model describes the encoding of an unobserved brain state in multiscale spike-field activity. It also allows for regime-switching dynamics using an unobserved regime state that dictates the dynamical and encoding parameters at every time-step. We also design the associated switching multiscale inference methods that estimate both the unobserved regime and brain states from simultaneous spike-field activity.Main results.We validate the methods in both extensive numerical simulations and prefrontal spike-field data recorded in a monkey performing saccades for fluid rewards. We show that these methods can successfully combine the spiking and field potential observations to simultaneously track the regime and brain states accurately. Doing so, these methods lead to better state estimation compared with single-scale switching methods or stationary multiscale methods. Also, for single-scale linear Gaussian observations, the new switching smoother can better generalize to diverse system settings compared to prior switching smoothers.Significance.These modeling and inference methods effectively incorporate both regime-detection and multiscale observations. As such, they could facilitate investigation of latent switching neural population dynamics and improve future brain-machine interfaces by enabling inference in naturalistic scenarios where regime-dependent multiscale activity and behavior arise.

Rate of Unplanned Vitrectomies in Femtosecond Laser-Assisted Cataract Surgery Compared to Conventional Phacoemulsification.

PURPOSE: To assess the rate of unplanned vitrectomies with femtosecond laser-assisted cataract surgery (FLACS) compared to conventional phacoemulsification at a single, high-volume, multi-surgeon ambulatory surgical center. METHODS: Retrospective, comparative study of phacoemulsification cases performed in a community-based ambulatory surgical center. A chart review of 2,480 consecutive FLACS procedures performed by 30 surgeons and 36,865 consecutive conventional phacoemulsification surgeries performed by 47 surgeons was conducted. The rate of unplanned vitrectomies was evaluated. In eyes with unplanned vitrectomy, the rates of posterior capsule tear, anterior capsule tear, and zonular dehiscence were analyzed. The unplanned vitrectomy rates between early and late FLACS cases were compared using thresholds of greater than 10, 20, and 50 cases. RESULTS: There were 230 (0.62%) and 16 (0.65%) unplanned vitrectomies in the conventional phacoemulsification and FLACS groups, respectively (P = .89). Of the cases requiring unplanned vitrectomy, posterior capsule tear, anterior capsule tear, and zonular dehiscence rates (conventional phacoemulsification vs FLACS) were 70.9% versus 56.3%, 8.3% versus 12.5%, and 20.9% versus 31.3%, respectively (P = .35). The difference in unplanned vitrectomy rates between early and late cases did not reach statistical significance in any threshold group. CONCLUSIONS: FLACS had a rate of unplanned vitrectomy comparable to conventional phacoemulsification. There was no statistically significant difference in unplanned vitrectomy rates between early and late FLACS cases suggesting a short learning curve. [J Refract Surg. 2018;34(9):610-614.].

Assessment of surgeon experience with femtosecond laser-assisted cataract surgery.

PURPOSE: To evaluate the collective user experience with an image-guided femtosecond laser (FSL) for cataract surgery in a high-volume, multi-surgeon, ambulatory surgical center. SUBJECTS AND METHODS: A detailed online survey was distributed to all surgeons in a single ambulatory surgical center who had performed cataract surgery using a FSL since its acquisition in December 2012. Information collected included the number of cases performed, typical surgical techniques and parameters, satisfaction with individual features of the laser (rated on a scale from 1=completely unsatisfied to 10=extremely satisfied) and commentary on ease of use and suggested improvements. RESULTS: Seventeen of 30 surgeons (56.7%) completed the survey, representing a case volume of 1,967 eyes. Fourteen surgeons (82.4%) felt they required ≤10 cases with the FSL to operate with the same safety and control as in standard phacoemulsification surgery. Satisfaction was highest for capsulotomies, lens fragmentation, lens softening, arcuate incisions and the graphic user interface (mean scores 9.4, 8.7, 8.7, 7.2 and 8.9, respectively). Preferred capsulotomy diameter was 4.8-5.2 mm (64.7% of respondents). About half (52.9%) of respondents centered the capsulotomy on the pupil and the other 47.1% centered the capsulotomy using optical coherence tomography. Most respondents (81.3%) preferred transepithelial arcuate incisions compared to intrastromal incisions. Satisfaction was lowest with FSL-created, main, clear corneal incisions and paracenteses (mean scores 4.4 and 4.2, respectively). CONCLUSION: Laser-assisted cataract surgery has a short learning curve and a high rate of user satisfaction. Further software and hardware development is warranted to improve user satisfaction with peripheral and clear corneal incisions.

A novel nomogram for the treatment of astigmatism with femtosecond-laser arcuate incisions at the time of cataract surgery.

PURPOSE: The purpose of this study was to determine an arcuate incision (AI) nomogram to treat astigmatism during femtosecond laser-assisted cataract surgery. METHODS: This is a retrospective, cohort study. Femtosecond laser (FSL)-assisted transepithelial AIs were created at a 9.0 mm optical zone, 80% depth, centered on the limbus. We modified the manual Donnenfeld limbal relaxing incision nomogram to 70% for with-the-rule (WTR), 80% for oblique (OBL), and 100% for against-the-rule (ATR) astigmatism. The correction index (CI) equaled AI-induced astigmatism/target-induced astigmatism. Measures included preoperative keratometric corneal cylinder (Pre Kcyl), postoperative Kcyl (Post Kcyl), and postoperative residual refractive astigmatism (Post RRA). RESULTS: Mean Pre Kcyl and 1-2 months Post RRA in 161 eyes of 116 patients were 0.626±0.417 diopters (D) (range 0.5-2 D), and 0.495±0.400 D (range 0-1.5 D), respectively. Mean absolute astigmatic changes (Pre Kcyl-Post Kcyl) without accounting for axis change in the WTR, ATR, and OBL groups were 0.165±0.383 D (P<0.001), 0.374±0.536 D (P<0.001), and 0.253±0.416 D (P=0.02), respectively. Mean absolute astigmatic changes using RRA as the postoperative measurement (Pre Kcyl-Post RRA) without accounting for axis change were 0.440±0.461 D (P<0.001), 0.238±0.571 D (P<0.05), 0.154±0.450 (P=0.111) in WTR, ATR, and OBL groups, respectively. CIs for WTR, ATR, and OBL were 0.53, 1.01, and 0.95, respectively. There were no intraoperative or postoperative complications related to the AIs. CONCLUSION: Transepithelial FSL-AIs using the modified Donnenfeld nomogram show potential for management of mild to moderate corneal astigmatism. An increase in the magnitude or reduction of the optical zone size for the treatment of WTR and ATR astigmatism for this nomogram may further improve refractive accuracy.

Defining 10-2 visual field progression criteria: exploratory and confirmatory factor analysis using pointwise linear regression.

PURPOSE: To test different visual field progression criteria using trend analysis in a glaucoma population followed with long sequences of 10-2 tests as a first attempt to understand and document rates of progression in the central field. DESIGN: Retrospective cohort study. PARTICIPANTS: We included 146 eyes of 146 patients with established glaucoma. METHODS: Pointwise linear regression analysis using the methods of ordinary least squares was performed on the 68 test locations of the 10-2 visual field sequences. Threshold sensitivities at each test location were plotted as the dependent variable against follow-up time as the independent variable. Statistically significant progression or improvement of a visual field test point was defined if its regression slope measured ≤-1.0 dB/year or ≥+1.0 dB/year, respectively, at P<0.01. We explored sets of criteria to define visual field progression, generating a hypothetical sensitivity (progression), specificity (improvement), and progression-to-improvement ratio (PIR) for each criterion. The criterion with the highest PIR was deemed the one with best performance. Latent class analysis (LCA) was used to determine visual field sectors with highest inter-correlation. MAIN OUTCOME MEASURES: The performance of different visual field progression criteria to detect fast rates of mean deviation (MD) change. RESULTS: Median baseline 10-2 MD value was -12.0 dB (interquartile range [IQR], -6.7 to -17.8 dB), and the median rate of 10-2 MD change over time was -0.38 dB/year (IQR, -0.07 to -0.77 dB/year). The highest PIR was obtained with the progression criterion requiring at least 3 test points located in the same LCA-derived 10-2 visual field sector progressing faster than -1.0 dB/year at P<0.01. This criterion was further validated for content and convergence. CONCLUSIONS: This is the first study to investigate progression criteria for 10-2 visual fields using rates of change and to test their performance and validity. These findings may be useful to improve the monitoring of patients with glaucoma at different levels of functional loss and to develop new perimetric algorithms that scrutinize specific visual field locations for a more accurate detection of progression.

Risk calculation variability over time in ocular hypertensive subjects.

PURPOSE: To investigate the longitudinal variability of glaucoma risk calculation in ocular hypertensive (OHT) subjects. METHODS: We reviewed the charts of untreated OHT patients followed in a glaucoma referral practice for a minimum of 60 months. Clinical variables collected at baseline and during follow-up included age, central corneal thickness (CCT), intraocular pressure (IOP), vertical cup-to-disc ratio (VCDR), and visual field pattern standard deviation (VFPSD). These were used to calculate the 5-year risk of conversion to primary open-angle glaucoma (POAG) at each follow-up visit using the Ocular Hypertension Treatment Study and European Glaucoma Prevention Study calculator (http://ohts.wustl.edu/risk/calculator.html). We also calculated the risk of POAG conversion based on the fluctuation of measured variables over time assuming the worst case scenarios (final age, highest PSD, lowest CCT, highest IOP, and highest VCDR) and best case scenarios (baseline age, lowest PSD, highest CCT, lowest IOP, and lowest VCDR) for each patient. Risk probabilities (%) were plotted against follow-up time to generate slopes of risk change over time. RESULTS: We included 27 untreated OHT patients (54 eyes) followed for a mean of 98.3±18.5 months. Seven individuals (25.9%) converted to POAG during follow-up. The mean 5-year risk of conversion for all patients in the study group ranged from 2.9% to 52.3% during follow-up. The mean slope of risk change over time was 0.37±0.81% increase/y. The mean slope for patients who reached a POAG endpoint was significantly greater than for those who did not (1.3±0.78 vs. 0.042±0.52%/y, P<0.01). In each patient, the mean risk of POAG conversion increased almost 10-fold when comparing the best case scenario with the worst case scenario (5.0% vs. 45.7%, P<0.01). CONCLUSIONS: The estimated 5-year risk of conversion to POAG among untreated OHT patients varies significantly during follow-up, with a trend toward increasing over time. Within the same individual, the estimated risk can vary almost 10-fold based on the variability of IOP, CCT, VCDR, and VFPSD. Therefore, a single risk calculation measurement may not be sufficient for accurate risk assessment, informed decision-making by patients, and physician treatment recommendations.

Morphogenetic and cellular movements that shape the mouse cerebellum; insights from genetic fate mapping.

We used the cerebellum as a model to study the morphogenetic and cellular processes underlying the formation of elaborate brain structures from a simple neural tube, using an inducible genetic fate mapping approach in mouse. We demonstrate how a 90 degrees rotation between embryonic days 9 and 12 converts the rostral-caudal axis of dorsal rhombomere 1 into the medial-lateral axis of the wing-like bilateral cerebellar primordium. With the appropriate use of promoters, we marked specific medial-lateral domains of the cerebellar primordium and derived a positional fate map of the murine cerebellum. We show that the adult medial cerebellum is produced by expansion, rather than fusion, of the thin medial primordium. Furthermore, ventricular-derived cells maintain their original medial-lateral coordinates into the adult, whereas rhombic lip-derived granule cells undergo lateral to medial posterior transverse migrations during foliation. Thus, we show that progressive changes in the axes of the cerebellum underlie its genesis.