Alzheimer's disease (AD) and frontotemporal dementia (FTD) are both progressive neurodegenerative diseases that severely affect the quality of life for patients. Studies involving the use of electroencephalogram (EEG) to diagnose such conditions have recently gained popularity due to the ease of collection of EEG and the application of computer algorithms for its analysis. In this paper, we propose to use a time-frequency analysis method, namely Variational Mode Decomposition (VMD), and 2D Convolutional Neural Network (CNN) to classify 13-minute-long EEG recordings as belonging to control subjects or patients with AD or FTD. We achieve an accuracy of 93.7% for 3-class classification and accuracies of 94.5% and 96.1% for binary classification of FTD vs control subjects and AD vs control subjects, respectively.

The outbreak of the 2019 novel Coronavirus (COVID-19) has affected millions of people worldwide. The structural manifestations of COVID-19 can be visualized on radiographic images of the chest, including computerized tomography (CT) and X-ray. Advancements in machine learning have led to the development of transfer learning models as tools for faster and more accurate diagnoses of COVID-19 when compared to the reverse transcription-polymerase chain reaction (RT-PCR) test. Transfer learning utilizes pre-trained convolutional neural network (CNN) models to classify images from an unseen dataset. We analyzed the performance of six different transfer learning architectures on their ability to classify 2482 single slice chest CT images from 60 COVID-19 positive patients and 60 COVID-19 negative patients. The best performing model was InceptionV3, with 98.71 ± 1.43% validation accuracy, 97.64 ± 3.06% specificity, 99.76 ± 0.36% sensitivity, 97.73 ± 2.78% precision, an F1 score of 98.74 ± 1.37% and an AUC of 0.9997 ± 0.0003. These preliminary results suggest that applying an InceptionV3 transfer learning-based framework on chest CT scans is a very effective method for the classification of COVID-19 patients. However, the diagnostic capacity of machine learning models would greatly improve with the introduction of additional classification groups, such as disease severity or other lung diseases. Therefore, we hope to adapt a machine learning architecture to receive multiple inputs and apply it to multi-modal data types, such as imaging, demographic and clinical. In addition to chest images, other inputs would include sex, age, comorbidities, or clinical lab results that can potentially improve COVID-19 prediction. An explicit comparison between classification on imaging data, clinical data, and a combination of the modalities will provide great insight into the clinical relevance of machine learning methods.

As new variants of the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) continuously evolve and mutate, it is vital to understand how novel and emerging variants affect public health outcomes. Our understanding of COVID-19 can have significant impacts on how we approach measures to mitigate the virus’s deleterious effects on the world. Therefore, in this study, we aim to 1) quantify the relationships between county-characteristics, such as the proportion of residents vaccinated, and certain county-level health outcomes related to COVID-19 such as: case counts, death counts, positivity rates, infection rates, ICU occupancy levels, hospitalizations, and the proportion of ICU admissions due to COVID-19. We also aim to 2) compare these relationships across three different time periods - two periods where Delta was the dominant strain of the U.S. and one period where Omicron was the dominant strain of the U.S. In this study, we used multiple regression to measure the strength of relationships between healthcare outcomes and county characteristics from June 20, 2021 to March 19, 2022, which span across three time periods. The first two time periods, June 20, 2021 to September 18, 2021 and September 19, 2021 to December 18, 2021, are when Delta was dominant (> 50% of cases) in the U.S. and the third period, December 19, 2021 to March 19, 2022, is during Omicron’s dominance (up to March 19, 2022).

Concerns regarding Coronavirus disease 2019 (COVID-19) disparities among the vulnerable populations have been studied widely and extend to vaccine allocation. With the current status of the pandemic, emergence of new variants, and recommendations toward the administration of third and fourth booster shots, understanding the population-level vaccine coverage status is critical for achieving equity in public health outcomes. In a previous population-level study focused on California counties, using the data from U.S. Census Bureau, Agency for Toxic Substances and Disease Registry (ATSDR), California Health & Human Services Agency (CHHS), and Centers for Disease Control and Prevention (CDC), we investigated the scope of vaccine inequity. In this research, we modeled the trends of vaccination using the Social Vulnerability Index (SVI) and its four components, including "socioeconomic status," "household composition and disability," "minority status and language," and "housing type and transportation." Growth velocity and anticipated maximum proportion of the vaccinated population were modeled based on 9228 data points of daily vaccination numbers from December 15, 2020, to May 23, 2021, SVI, and its four themes. We reported that although vaccination coverage was significantly lower in high vulnerability counties, they are estimated to achieve a higher proportion of vaccinated individuals eventually, particularly based on SVI Theme 3, while predicting that disparities in vaccination rates decrease over time.

Epilepsy is one of the most common neurological disorders, typically observed via seizure episodes. Epileptic seizures are commonly monitored through electroencephalogram (EEG) recordings due to their routine and low expense collection. The stochastic nature of EEG makes seizure identification via manual inspections performed by highly-trained experts a tedious endeavor, motivating the use of automated identification. The literature on automated identification focuses mostly on supervised learning methods requiring expert labels of EEG segments that contain seizures, which are difficult to obtain. Motivated by these observations, we pose seizure identification as an unsupervised anomaly detection problem. To this end, we employ the first unsupervised transformer-based model for seizure identification on raw EEG. We train an autoencoder involving a transformer encoder via an unsupervised loss function, incorporating a novel masking strategy uniquely designed for multivariate time-series data such as EEG. Training employs EEG recordings that do not contain any seizures, while seizures are identified with respect to reconstruction errors at inference time. We evaluate our method on three publicly available benchmark EEG datasets for distinguishing seizure vs. non-seizure windows. Our method leads to significantly better seizure identification performance than supervised learning counterparts, by up to 16% recall, 9% accuracy, and 9% Area under the Receiver Operating Characteristics Curve (AUC), establishing particular benefits on highly imbalanced data. Through accurate seizure identification, our method could facilitate widely accessible and early detection of epilepsy development, without needing expensive label collection or manual feature extraction.

Visualization is a critical component of neuroimaging, and how to best view data that is naturally three dimensional is a long standing question in neuroscience. Many approaches, programs, and techniques have been developed specifically for neuroimaging. However, exploration of 3D information through a 2D screen is inherently limited. Many neuroscientific researchers hope that with the recent commercialization and popularization of VR, it can offer the next-step in data visualization and exploration. Neuro Imaging in Virtual Reality (NIVR), is a visualization suite that employs various immersive visualizations to represent neuroimaging information in VR. Some established techniques, such as raymarching volume visualization, are paired with newer techniques, such as near-field rendering, to provide a broad basis of how we can leverage VR to improve visualization and navigation of neuroimaging data. Several of the neuroscientific visualization approaches presented are, to our knowledge, the first of their kind. NIVR offers not only an exploration of neuroscientific data visualization, but also a tool to expose and educate the public regarding recent advancements in the field of neuroimaging. By providing an engaging experience to explore new techniques and discoveries in neuroimaging, we hope to spark scientific interest through a broad audience. Furthermore, neuroimaging offers deep and expansive datasets; a single scan can involve several gigabytes of information. Visualization and exploration of this type of information can be challenging, and real-time exploration of this information in VR even more so. NIVR explores pathways which make this possible, and offers preliminary stereo visualizations of these types of massive data.

The USC Stevens Neuroimaging and Informatics Institute in the Laboratory of Neuro Imaging (http://loni.usc.edu) has the largest collection/repository of neuroanatomical MRI scans in the world and is at the forefront of both brain imaging and data storage/processing technology. One of our workflow processes involves algorithmic segmentation of MRI scans into labeled anatomical regions (using FreeSurfer, currently the best software for this purpose). This algorithm is imprecise, and users must tediously correct errors manually by using a mouse and keyboard to edit individual MRI slices at a time. We demonstrate preliminary work to improve efficiency of this task by translating it into 3 dimensions and utilizing virtual reality user interfaces to edit multiple slices of data simultaneously.