Search for: All records

We present amulti-modal fiber array snapshot technique (M-FAST)based on an array of 96 compact cameras placed behind a primary objective lens and a fiber bundle array. Our technique is capable of large-area, high-resolution, multi-channel video acquisition. The proposed design provides two key improvements to prior cascaded imaging system approaches: a novel optical arrangement that accommodates the use of planar camera arrays, and a new ability to acquire multi-modal image data acquisition. M-FAST is a multi-modal, scalable imaging system that can acquire snapshot dual-channel fluorescence images as well as differential phase contrast measurements over a large 6.59 mm × 9.74 mm field-of-view at 2.2-μm center full-pitch resolution.

 

Abstract Background Diabetic retinopathy (DR) is a leading cause of blindness in American adults. If detected, DR can be treated to prevent further damage causing blindness. There is an increasing interest in developing artificial intelligence (AI) technologies to help detect DR using electronic health records. The lesion-related information documented in fundus image reports is a valuable resource that could help diagnoses of DR in clinical decision support systems. However, most studies for AI-based DR diagnoses are mainly based on medical images; there is limited studies to explore the lesion-related information captured in the free text image reports. Methods In this study, we examined two state-of-the-art transformer-based natural language processing (NLP) models, including BERT and RoBERTa, compared them with a recurrent neural network implemented using Long short-term memory (LSTM) to extract DR-related concepts from clinical narratives. We identified four different categories of DR-related clinical concepts including lesions, eye parts, laterality, and severity, developed annotation guidelines, annotated a DR-corpus of 536 image reports, and developed transformer-based NLP models for clinical concept extraction and relation extraction. We also examined the relation extraction under two settings including ‘gold-standard’ setting—where gold-standard concepts were used–and end-to-end setting. Results For concept extraction, the BERT model pretrained with the MIMIC III dataset achieve the best performance (0.9503 and 0.9645 for strict/lenient evaluation). For relation extraction, BERT model pretrained using general English text achieved the best strict/lenient F1-score of 0.9316. The end-to-end system, BERT_general_e2e, achieved the best strict/lenient F1-score of 0.8578 and 0.8881, respectively. Another end-to-end system based on the RoBERTa architecture, RoBERTa_general_e2e, also achieved the same performance as BERT_general_e2e in strict scores. Conclusions This study demonstrated the efficiency of transformer-based NLP models for clinical concept extraction and relation extraction. Our results show that it’s necessary to pretrain transformer models using clinical text to optimize the performance for clinical concept extraction. Whereas, for relation extraction, transformers pretrained using general English text perform better. 

Crowd mobility prediction, in particular, forecasting flows at and transitions across different locations, is essential for crowd analytics and management in spacious environments featured with large gathering. We propose GAEFT, a novel crowd mobility analytics system based on the multi-task graph attention neural network to forecast crowd flows (inflows/outflows) and transitions. Specifically, we leverage the collective and sanitized campus Wi-Fi association data provided by our university information technology service and conduct a relatable case study. Our comprehensive data analysis reveals the important challenges of sparsity and skewness, as well as the complex spatio-temporal variations within the crowd mobility data. Therefore, we design a novel spatio-temporal clustering method to group Wi-Fi access points (APs) with similar transition features, and obtain more regular mobility features for model inputs. We then propose an attention-based graph embedding design to capture the correlations among the crowd flows and transitions, and jointly predict the AP-level flows as well as transitions across buildings and clusters through a multi-task formulation. Extensive experimental studies using more than 28 million association records collected during 2020-2021 academic year validate the excellent accuracy of GAEFT in forecasting dynamic and complex crowd mobility.