A Large Language Modelling Deep Learning Framework for the Next Pandemic (preprint)

Deep neural networks have been integrated into the whole clinical decision procedure which can improve the efficiency of diagnosis and alleviate the heavy workload of physicians. Typical applications include 1) medical report generation, 2) disease classification, and 3) survival prediction. Since most neural networks are supervised, their quality heavily depends on the volume and quality of available labels. However, for novel diseases, e.g., new pandemics or new variants, there are few existing labels. In addition, the acquisition of new pandemic cases to collect sufficient labels for training is time-consuming and is typically unavailable at the early stage. To prepare neural networks for the next pandemic, in this paper, we propose a large language model - Unsupervised Learning from Unlabelled Medical Images and Text (ULUMIT) framework, which can learn broad medical knowledge (e.g., image understanding, text semantics, and clinical phenotypes) from unlabelled data. As a result, when encountering new pandemics, our framework can be rapidly deployed and easily adapted to them with extremely limited labels. Furthermore, ULUMIT supports medical data across visual modality (e.g., chest X-ray and CT) and textual modality (e.g., medical report and free-text clinical note); therefore, it can be used for any clinical task that involves both visual and textual medical data. We demonstrate the effectiveness of our ULUMIT by showing how it would perform using the COVID-19 pandemic ``in replay''. In particular, in the retrospective setting, we test the model on the early COVID-19 datasets; and in the prospective setting, we test the model on the new variant COVID-19-Omicron. The experiments are conducted on 1) three kinds of input medical data, image-only, text-only, and image-text; 2) three kinds of downstream tasks, medical reporting, diagnosis, and prognosis; 3) five public COVID-19 datasets; and 4) three different languages, i.e., English, Chinese, and Spanish. All experiments consistently show that our framework can make accurate and robust COVID-19 decision-support tasks with little labelled data (such as considering information from only one patient), providing an impact on medical data analysis during the early stage of the next pandemic. Besides COVID-19, our framework can be applied to identify 14 common thorax diseases and tuberculosis across five additional public datasets, demonstrating its robustness in generalization and transferability. In brief, our framework achieves state-of-the-art performances on ten datasets.

Non-linear feature analysis of public emotion evolution for online teaching during the COVID-19 pandemic

Purpose Nowadays, the breakout of the COVID-19 pandemic has caused an important change in teaching models. The emotional experience of this change has an important impact on online teaching. This paper aims to explore its time evolution characteristics and provide reference for the development of online teaching in the post epidemic era. Design/methodology/approach The article firstly crawls the online teaching-related comment text data on Zhihu platform and performs emotional calculation to obtain a one-dimensional time series of daily average emotional values. Then, by using non-linear time-series analysis, this paper reconstructs the daily average emotion value time series in high-dimensional phase space, calculates the maximum Lyapunov exponent and correlation dimension and finally, explores the feature patterns through recurrence plot and recurrence quantification analysis. Findings It was found that the sequence has typical non-linear chaotic characteristics;its correlation dimension indicates that it contains obvious fractal characteristics;the public emotional evolution shows a cyclical rise and fall. By text mining and temporal evolution analysis, this paper explores the evolution law over chronically of the daily average emotion value time series, provides feasible strategies to improve students' online learning experience and quality and continuously optimizes this new teaching model in the era of pandemic. Originality/value Based on social knowledge sharing platform of Q&A, this paper models and analyzes users interaction data under online teaching-related topics. This paper explores the evolution law over a long time period of the daily average emotion value time series using text mining and temporal evolution analysis. It then offers workable solutions to enhance the quality and experience of students' online learning, and it continuously improves this new teaching model in the age of pandemics.

Sub10 μm macroporous aramid substrates with a hierarchically structured interface for organic solvent nanofiltration

Compared with traditional methods for elaborately tailoring the active layer of thin-film composite (TFC) membranes, this study focused on building novel substrates for potential applications in developing organic solvent nanofiltration (OSN) membranes. One kind of “three-parts” hierarchically structured interface with nanospheres and macro surface pores was successfully prepared via Michael addition and Schiff's base (M&S) reactions between N-(2-aminoethyl)-3-aminopropyl triethoxysilane (NAE-A) and glycerite (a natural polyphenol) in the aramid substrate. The thickness was reduced to sub10 μm with the aid of the high-speed spin coating process coupling nonsolvent-induced phase separation (HSSC-co-NIPS) method. The composition characterization results demonstrated the introduction of glycerite, and the reactions occurred in/on the substrate. Scanning electron microscopy (SEM) images clearly showed that the sub10 μm substrate contained a “three-level” hierarchically structured interface that included: (1) a “stalk-like” structure;(2) glycerite-NAE-A silane (GNAS) nanospheres;and (3) the substrate surface. These phenomena also resulted in better surface hydrophilicity. All types of organic solvents, including harsh solvents, such as tetrahydrofuran (THF) and N,N-dimethylformamide (DMF), had stable permeability within the substrate, as did apolar n-hexane and isopar™ G. Therefore, the as-prepared TFC OSN membrane, which had an extremely short polymerization time, retained broad-spectrum solvent stability and had the highest solvent permeance in acetonitrile (24.5 ± 0.4 L m−2 h−1·bar−1). In addition, the resultant TFC membrane almost completely rejected the popular macrolide antibiotic azithromycin (AZM, 748.98 g mol−1) in ethanol, which is used to treat COVID-19.