Sean Wu’s research while affiliated with Pepperdine University and other places

Perimetric measurements provide insight into a patient's peripheral vision and day-to-day functioning and are the main outcome measure for identifying progression of visual damage from glaucoma. However, visual field data can be noisy, exhibiting high variance, especially with increasing damage. In this study, we demonstrate the utility of self-supervised deep learning in denoising visual field data from over 4000 patients to enhance its signal-to-noise ratio and its ability to detect true glaucoma progression. We deployed both a variational autoencoder (VAE) and a masked autoencoder to determine which self-supervised model best smooths the visual field data while reconstructing salient features that are less noisy and more predictive of worsening disease. Our results indicate that including a categorical p-value at every visual field location improves the smoothing of visual field data. Masked autoencoders led to cleaner denoised data than previous methods, such as variational autoencoders. A 4.7% increase in detection of progressing eyes with pointwise linear regression (PLR) was observed. The masked and variational autoencoders' smoothed data predicted glaucoma progression 2.3 months earlier when p-values were included compared to when they were not. The faster prediction of time to progression (TTP) and the higher percentage progression detected support our hypothesis that masking out visual field elements during training while including p-values at each location would improve the task of detection of visual field progression. Our study has clinically relevant implications regarding masking when training neural networks to denoise visual field data, resulting in earlier and more accurate detection of glaucoma progression. This denoising model can be integrated into future models for visual field analysis to enhance detection of glaucoma progression.

3D reconstruction of biplane cerebral angiograms remains a challenging, unsolved research problem due to the loss of depth information and the unknown pixelwise correlation between input images. The occlusions arising from only two views complicate the reconstruction of fine vessel details and the simultaneous addressing of inherent missing information. In this paper, we take an incremental step toward solving this problem by reconstructing the corresponding 2D slice of the cerebral angiogram using biplane 1D image data. We developed a coordinate-based neural network that encodes the 1D image data along with a deterministic Fourier feature mapping from a given input point, resulting in a slice reconstruction that is more spatially accurate. Using only one 1D row of biplane image data, our Fourier feature network reconstructed the corresponding volume slices with a peak signal-to-noise ratio (PSNR) of 26.32 ± 0.36, a structural similarity index measure (SSIM) of 61.38 ± 1.79, a mean squared error (MSE) of 0.0023 ± 0.0002, and a mean absolute error (MAE) of 0.0364 ± 0.0029. Our research has implications for future work aimed at improving backprojection-based reconstruction by first examining individual slices from 1D information as a prerequisite.

Open-source LLMs have shown great potential as fine-tuned chatbots, and demonstrate robust abilities in reasoning and surpass many existing benchmarks. Retrieval-Augmented Generation (RAG) is a technique for improving the performance of LLMs on tasks that the models weren't explicitly trained on, by leveraging external knowledge databases. Numerous studies have demonstrated the effectiveness of RAG to more successfully accomplish downstream tasks when using vector datasets that consist of relevant background information. It has been implicitly assumed by those in the field that if adversarial background information is utilized in this context, that the success of using a RAG-based approach would be nonexistent or even negatively impact the results. To address this assumption, we tested several open-source LLMs on the ability of RAG to improve their success in answering multiple-choice questions (MCQ) in the medical subspecialty field of Nephrology. Unlike previous studies, we examined the effect of RAG in utilizing both relevant and adversarial background databases. We set up several open-source LLMs, including Llama 3, Phi-3, Mixtral 8x7b, Zephyr$\beta$, and Gemma 7B Instruct, in a zero-shot RAG pipeline. As adversarial sources of information, text from the Bible and a Random Words generated database were used for comparison. Our data show that most of the open-source LLMs improve their multiple-choice test-taking success as expected when incorporating relevant information vector databases. Surprisingly however, adversarial Bible text significantly improved the success of many LLMs and even random word text improved test taking ability of some of the models. In summary, our results demonstrate for the first time the countertintuitive ability of adversarial information datasets to improve the RAG-based LLM success.

3D reconstruction of biplane cerebral angiograms remains a challenging, unsolved research problem due to the loss of depth information and the unknown pixelwise correlation between input images. The occlusions arising from only two views complicate the reconstruction of fine vessel details and the simultaneous addressing of inherent missing information. In this paper, we take an incremental step toward solving this problem by reconstructing the corresponding 2D slice of the cerebral angiogram using biplane 1D image data. We developed a coordinate-based neural network that encodes the 1D image data along with a deterministic Fourier feature mapping from a given input point, resulting in a slice reconstruction that is more spatially accurate. Using only one 1D row of biplane image data, our Fourier feature network reconstructed the corresponding volume slices with a peak signal-to-noise ratio (PSNR) of 26.32±0.36, a structural similarity index measure (SSIM) of 61.38±1.79, a mean squared error (MSE) of 0.0023±0.0002, and a mean absolute error (MAE) of 0.0364±0.0029. Our research has implications for future work aimed at improving backprojection-based reconstruction by first examining individual slices from 1D information as a prerequisite.

Hydrofoil surfing is a distinct subclass of surfing, where riders navigate unbroken waves while elevated above the water’s surface through the use of a hydrofoil affixed to the underside of the surfboard. Foiling has been popularized due to an increase in riding distance capabilities. In particular, the downwind technique allows riders to surf on open ocean wind swells taking advantage of the hydrofoil’s hydrodynamic ability to generate lift. However, the downwind technique in hydrofoil surfing maintains a high barrier to entry for novice users due to the real-time task of visually identifying and transitioning between the frequently changing wind waves. Thus, there arises a need for a user-friendly system capable of prompt identification and assessment of wave quality for foiling. To meet this demand, we propose Foil-Net, a wave classification tool developed for hydrofoil surfing. Foil-Net leverages a combination of an autoencoder [10] and convolutional neural network (CNN) [11] classifier to categorize and index waves based on their suitability for hydrofoil surfing.

Accurately capturing the 3D geometry of the brain’s blood vessels is critical in helping neuro-interventionalists to identify and treat neurovascular disorders, such as stroke and aneurysms. Currently, the gold standard for obtaining a 3D representation of angiograms is through the process of 3D rotational angiography, a timely process requiring expensive machinery, which is also associated with high radiation exposure to the patient. In this research, we propose a new technique for reconstructing 3D volumes from 2D angiographic images, thereby reducing harmful X-ray radiation exposure. Our approach involves parameterizing the input data as a back-projected noisy volume from the images, which is then fed into a 3D denoising autoencoder. Through this method, we have achieved clinically relevant reconstructions with varying amounts of 2D projections from 49 patients. Additionally, our 3D denoising autoencoder outperformed previous generative models in biplane reconstruction by 15.51% for intersection over union (IOU) and 3.5% in pixel accuracy due to keeping a semi-accurate input with back projection. This research highlights the significant role of back-projection in achieving relative visual correspondence in the input space to reconstruct 3D volumes from 2D angiograms. This approach has the potential to be deployed in future neurovascular surgery, where 3D volumes of the patient’s brain blood vessels can be visualized with less X-ray radiation exposure time.

Purpose: Predict central 10° global and local visual field (VF) measurements from macular optical coherence tomography (OCT) volume scans with deep learning (DL). Methods: This study included 1121 OCT volume scans and 10-2 VFs from 289 eyes (257 patients). Macular scans were used to estimate 10-2 VF mean deviation (MD), threshold sensitivities (TS), and total deviation (TD) values at 68 locations. A three-dimensional (3D) convolutional neural network based on the 3D DenseNet121 architecture was used for prediction. We compared DL predictions to those from baseline linear models. We carried out 10-fold stratified cross-validation to optimize generalizability. The performance of the DL and baseline models was compared based on correlations between ground truth and predicted VF measures and mean absolute error (MAE; ground truth - predicted values). Results: Average (SD) MD was -9.3 (7.7) dB. Average (SD) correlations between predicted and ground truth MD and MD MAE were 0.74 (0.09) and 3.5 (0.4) dB, respectively. Estimation accuracy deteriorated with worsening MD. Average (SD) Pearson correlations between predicted and ground truth TS and MAEs for DL and baseline model were 0.71 (0.05) and 0.52 (0.05) (P < 0.001) and 6.5 (0.6) and 7.5 (0.5) dB (P < 0.001), respectively. For TD, correlation (SD) and MAE (SD) for DL and baseline models were 0.69 (0.02) and 0.48 (0.05) (P < 0.001) and 6.1 (0.5) and 7.8 (0.5) dB (P < 0.001), respectively. Conclusions: Macular OCT volume scans can be used to predict global central VF parameters with clinically relevant accuracy. Translational relevance: Macular OCT imaging may be used to confirm and supplement central VF findings using deep learning.