Multiple Criterion and Multiple Stimulus Signal Detection Theory Analysis of Corneal Painful and Cool Pneumatic Stimuli.

Purpose: To evaluate the detectability of pneumatic corneal stimuli and response bias using multi-stimuli multi-criterion signal detection theory (MSDT). Methods: Thirty-six participants were recruited using convenience sampling. A Waterloo Belmonte esthesiometer was used to deliver cold, mechanical, and chemical stimuli to the center of the cornea at three separate study visits. The stimulus type was assigned randomly to each visit at the start of the study. The threshold (baseline for detection theory experiment) for the assigned stimulus type was obtained using the ascending method of limits. In the cold and mechanical MSDT experiments, 100 trials (80 signal (20 each for 4 intensities) and 20 catch trials) were presented in randomized order, and participants responded with a 5-point confidence rating to each trial. In the chemical MSDT experiments, 50 trials (20 signal trials each for two intensities and 10 catch trials) were presented, and responses were provided using 4-point confidence ratings. Detection theory indices were calculated individually and as groups, which were then analyzed using mixed models and paired t-tests. Results: Detectability (da) and the area under the curve (Az) were significantly different between stimulus intensities within each stimulus type (all p < 0.001) but were not different between the stimulus types. Receiver operating characteristics (ROC) curves were separable between the scaled intensities for all stimulus types, and no overlaps were observed in the z-ROC space. The log-likelihood ratio (lnß) depended on stimulus intensity and psychophysical criterion for all stimulus types. Conclusion: It is feasible to use MSDT for analyzing ocular surface sensory processing and the theory provides insight into the possible bias associated with the use of pneumatic stimuli. With noxious and non-noxious pneumatic stimulation, detectability and criteria vary systematically with stimulus intensity, a result that cannot be derived using classical psychophysics and this highlights the importance of signal detection theory and its approaches in studying ocular surface pain and thermal processing.

Evaluation of Explainable Deep Learning Methods for Ophthalmic Diagnosis.

BACKGROUND: The lack of explanations for the decisions made by deep learning algorithms has hampered their acceptance by the clinical community despite highly accurate results on multiple problems. Attribution methods explaining deep learning models have been tested on medical imaging problems. The performance of various attribution methods has been compared for models trained on standard machine learning datasets but not on medical images. In this study, we performed a comparative analysis to determine the method with the best explanations for retinal OCT diagnosis. METHODS: A well-known deep learning model, Inception-v3 was trained to diagnose 3 retinal diseases - choroidal neovascularization (CNV), diabetic macular edema (DME), and drusen. The explanations from 13 different attribution methods were rated by a panel of 14 clinicians for clinical significance. Feedback was obtained from the clinicians regarding the current and future scope of such methods. RESULTS: An attribution method based on Taylor series expansion, called Deep Taylor, was rated the highest by clinicians with a median rating of 3.85/5. It was followed by Guided backpropagation (GBP), and SHapley Additive exPlanations (SHAP). CONCLUSION: Explanations from the top methods were able to highlight the structures for each disease - fluid accumulation for CNV, the boundaries of edema for DME, and bumpy areas of retinal pigment epithelium (RPE) for drusen. The most suitable method for a specific medical diagnosis task may be different from the one considered best for conventional tasks. Overall, there was a high degree of acceptance from the clinicians surveyed in the study.

Detectability and Bias Indices of Pneumatic Corneal Stimuli Using Signal Detection Theory.

Purpose: To evaluate the feasibility of using signal detection theory (SDT) in estimating criterion and detectability indices for corneal pneumatic stimuli and test corneal psychophysical data against linking hypotheses from nonprimate physiology using Bayesian analysis. Methods: Corneal pneumatic stimuli were delivered using the Waterloo Belmonte esthesiometer. Corneal thresholds were estimated in 30 asymptomatic participants and 1.5× threshold stimuli were used as signals (with 0.4 probability). There were 100-trial mechanical and cold stimulus experiments and 50-trial chemical experiments. Trials were demarcated auditorily and "yes" or "no" recorded after each trial. Cold stimulus experiments were conducted with 0.6 signal probability. Criterion (c), likelihood ratio (lnß), and d' were calculated from the yes-no responses. Results: Average d' was 0.59 ± 0.1, 1.65 ± 0.37, and 1.14 ± 0.3 units for cold, mechanical, and chemical stimuli, respectively. Bayes factors obtained using Bayesian analysis of variance mildly favored (BF10 = 1.55) differences between d's of the stimulus types, with no support for differences in criteria between stimulus types. Multiple comparisons of d' supported linking hypotheses based on nociception and nerve conductance theories. Conclusions: Our experiments are the first to demonstrate the feasibility of estimating SDT indices and test different hypotheses. The conservative strategy (reporting "no" more often) chosen by participants was anticipated due to relatively large proportion of catch trials. Translational Relevance: SDT when using pneumatic esthesiometry is vital to evaluate bias in responses of participants. Considering the varied forms of inherent noise in the corneal sensory system, SDT is critical to understand the sensory and decisional characteristics.

A Method to Calibrate the Carbon Dioxide (Chemical) Stimuli of Pneumatic Esthesiometer Externally.

PURPOSE: To determine the feasibility of using a portable carbon dioxide (CO2) sensor to calibrate a pneumatic esthesiometer and then to calibrate the chemical stimuli. METHODS: The chemical stimuli in ocular surface experiments are combinations of medical air and added CO2 (%CO2). These stimuli were calibrated using a portable CO2 sensor (COZIR CM-0041) and data logger, delivered for 100 seconds by using the Waterloo Belmonte esthesiometer. The distances between the sensor and esthesiometer tip were 0 (to measure feasibility), 3, 5, and 10 mm. In experiment I, 100% CO2 was tested using four different flow rates (50, 100, 150, and 200 mL/min) at three working distances. In experiment II, flow rates of 20 to 100 mL/min and concentrations of 20% to 100% CO2 were tested in 20 steps at 3 working distances. RESULTS: The CO2 sensor correctly reported the esthesiometer extremes of 0% and 100% CO2 when placed at the esthesiometer tip. There were progressive, systematic increases in concentrations reaching/reported by the sensor with increasing flow rates and nominal concentrations and progressive decreases in measurements with increases in working distance. CONCLUSIONS: CO2 concentrations in pneumatic esthesiometers can be calibrated and, as expected, vary with flow rate and distance, highlighting the importance of calibration and standardization of CO2 stimuli in these instruments. TRANSLATIONAL RELEVANCE: Calibrated CO2, a chemical sensory stimulus in humans, may be used in testing the surface of the eye as well as other membranes within which the CO2 can be dissolved (e.g., mucous) to produce an acidic stimulus.

Are the fluctuations in dynamic anterior surface aberrations of the human eye chaotic?

The purpose of the study is to measure chaos in dynamic anterior surface aberrations and examine how it varies between the eyes of an individual. Noninvasive tear breakup time and dynamic corneal surface aberrations were measured for two open-eye intervals of 15 s. The maximal Lyapunov exponent (MLE) was calculated to test the nature of the fluctuations of the dynamic anterior surface aberrations. The average MLE for total higher-order aberration (HOA) was found to be small (+0.0102±0.0072) µm/s. No significant difference in MLE was found between the eyes for HOA (t-test; p=0.131). Data analysis was carried out for individual Zernike coefficients, including vertical prism as it gives a direct measure of the thickness of the tear film over time. The results show that the amount of chaos was small for each Zernike coefficient and not significantly correlated between the eyes.