PurposeTo validate the effectiveness of an approach called batch-balanced focal loss (BBFL) in enhancing convolutional neural network (CNN) classification performance on imbalanced datasets.Materials and MethodsBBFL combines two strategies to tackle class imbalance: (1) batch-balancing to equalize model learning of class samples and (2) focal loss to add hard-sample importance to the learning gradient. BBFL was validated on two imbalanced fundus image datasets: a binary retinal nerve fiber layer defect (RNFLD) dataset (n = 7,258) and a multiclass glaucoma dataset (n = 7,873). BBFL was compared to several imbalanced learning techniques, including random oversampling (ROS), cost-sensitive learning, and thresholding, based on three state-of-the-art CNNs. Accuracy, F1-score, and the area under the receiver operator characteristic curve (AUC) were used as the performance metrics for binary classification. Mean accuracy and mean F1-score were used for multiclass classification. Confusion matrices, t-distributed neighbor embedding plots, and GradCAM were used for the visual assessment of performance.ResultsIn binary classification of RNFLD, BBFL with InceptionV3 (93.0% accuracy, 84.7% F1, 0.971 AUC) outperformed ROS (92.6% accuracy, 83.7% F1, 0.964 AUC), cost-sensitive learning (92.5% accuracy, 83.8% F1, 0.962 AUC), and thresholding (91.9% accuracy, 83.0% F1, 0.962 AUC) and others. In multiclass classification of glaucoma, BBFL with MobileNetV2 (79.7% accuracy, 69.6% average F1 score) outperformed ROS (76.8% accuracy, 64.7% F1), cost-sensitive learning (78.3% accuracy, 67.8.8% F1), and random undersampling (76.5% accuracy, 66.5% F1).ConclusionThe BBFL-based learning method can improve the performance of a CNN model in both binary and multiclass disease classification when the data are imbalanced.

With the rapid expansion of methods encompassed by the term gene therapy, new trials exploring the safety and efficacy of these methods are initiated more frequently. As a result, important questions arise pertaining the design of these trials and patient participation. One of the most important aspects of any clinical trial is the ability to measure the trial's outcome in a manner that will reflect the effect of the treatment and allow its quantification, whether the trial is aimed at preservation or restoration of retinal cells (photoreceptors and others), vision, or both. Here we will review the existing methods for quantification of trial outcomes, stressing the importance of assessing the participant's visual function and not just visual acuity. We will also describe the key considerations in trial design. Finally, as patient safety remains the primary concern in any trial participation, we will outline the key principles in that regard.

Background: With the clinical advances in the field of gene therapy, the development of objective measures of visual function of patients with inherited retinal dystrophies (IRDs) is of utmost importance. Here, we propose one such measure. Methods: We retrospectively analyzed data from a cohort of 194 eyes of 97 genetically diagnosed patients with retinitis pigmentosa (RP), the most common IRD, followed at the UPMC Vision Institute. The analyzed data included the reflectivity ratio (RR) of the retinal nerve fiber layer (RNFL) to that of the entire retina, visual acuity (VA) and the thickness of the retinal outer nuclear layer (ONL) and the RNFL. Results: There was a strong positive correlation between the RR and VA. Both VA and the RR were negatively correlated with disease duration; VA, but not the RR, was negatively correlated with age. The RR correlated with the ONL but not with the RNFL thickness or the intraocular pressure. Age, RR, disease duration and ONL thickness were found to be independent predictors of VA by multivariate analysis. Conclusion: The OCT RR could serve as an independent predictor of visual acuity, and by extension of retinal function, in genetically diagnosed RP patients. Such objective measures can be of great value in patient selection for therapeutic trials.

Purpose: To compare continuous wavelet transform (CWT) and short-time Fourier transform (STFT) analyses of the oscillatory potentials (OPs) of the dark-adapted full-field electroretinogram (FFERG). Methods: FFERG recordings from Royal College of Surgeons (RCS) and healthy Brown Norway (BN) rats, wild-type and Atoh7 enhancer-knockout (KO) mice, and patients with diabetic retinopathy and healthy controls were analyzed with STFT and with CWT employing the real Morlet wavelet, after 60 Hz high-pass and sharp 60 Hz notch filtering. Time- and frequency-localized OP features were quantified and the methods compared on detection of transient OP signals. Both were additionally validated using synthetic OP-like bursts of known frequency embedded in each cohort’s recorded noise. Results: CWT revealed distinct time–frequency patterns and disease-specific OP alterations across all rat, mouse, and human cohorts, with improved temporal localization and frequency resolution, whereas STFT merged transient features and was more susceptible to noise. CWT-derived dominant-frequency differences reached significance in all three models; the corresponding STFT difference reached significance only in rats (separate tests; methods not directly compared). In synthetic bursts, both methods recovered injected frequencies to within 1–2.5 Hz, but only CWT resolved two bursts separated by 15 ms. Conclusions: CWT employing the real Morlet wavelet provides a robust, physiologically relevant evaluation of OPs, with improved temporal localization and reduced spectral dispersion relative to STFT in clinical and experimental settings. Translational Relevance: CWT analysis may improve the characterization of inner retinal dysfunction by preserving transient OP time–frequency structure, as illustrated in preclinical rodent models and diabetic retinopathy patients.