Deep dynamic imputation of clinical time series for mortality prediction

Shi, Zhenkun; Wang, Sen; Yue, Lin; Li-xin, Pang; Zuo, Xianglin; Zuo, Wangmeng; Li, Xue

doi:10.1016/j.ins.2021.08.016

Cited by 16 publications

(6 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We model the clinical records using a BERT-based fine-tuning model, and then feed each clinical record of the patient into the textual part of our model separately, and then combine the resulting embeddings to obtain a final representation of the textual part of the data. We use a Long Short Term Memory (LSTM) network to model the pre-processed time series data and generate time series embeddings Shi et al (2021) . We then propose an improved fusion module for fusing features of single modalities, which partitions each modality into equal blocks of features on a channel and creates a joint representation for generating soft attention across feature blocks for each channel.…”

Section: Methodsmentioning

confidence: 99%

“…We use a Long Short Term Memory (LSTM) network to model the pre-processed time series data and generate time series embeddings Shi et al (2021). We then propose an improved fusion module for fusing features of single modalities, which partitions each modality into equal blocks of features on a channel and creates a joint representation for generating soft attention across feature blocks for each channel.…”

Section: The Overview Of Modelmentioning

confidence: 99%

See 1 more Smart Citation

Deep multi-modal intermediate fusion of clinical record and time series data in mortality prediction

Niu

Zhang²,

Peng³

et al. 2023

Front. Mol. Biosci.

View full text Add to dashboard Cite

In intensive care units (ICUs), mortality prediction is performed by combining information from these two sources of ICU patients by monitoring patient health. Respectively, time series data generated from each patient admission to the ICU and clinical records consisting of physician diagnostic summaries. However, existing mortality prediction studies mainly cascade the multimodal features of time series data and clinical records for prediction, ignoring thecross-modal correlation between the underlying features in different modal data. To address theseissues, we propose a multimodal fusion model for mortality prediction that jointly models patients’ time-series data as well as clinical records. We apply a fine-tuned Bert model (Bio-Bert) to the patient’s clinical record to generate a holistic embedding of the text part, which is then combined with the output of an LSTM model encoding the patient’s time-series data to extract valid features. The global contextual information of each modal data is extracted using the improved fusion module to capture the correlation between different modal data. Furthermore, the improved fusion module can be easily added to the fusion features of any unimodal network and utilize existing pre-trained unimodal model weights. We use a real dataset containing 18904 ICU patients to train and evaluate our model, and the research results show that the representations obtained by themodel can achieve better prediction accuracy compared to the baseline.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: The Overview Of Modelmentioning

confidence: 99%

Deep multi-modal intermediate fusion of clinical record and time series data in mortality prediction

Niu

Zhang²,

Peng³

et al. 2023

Front. Mol. Biosci.

View full text Add to dashboard Cite

show abstract

“…This can help in augmenting limited datasets and improving the robustness of predictive models trained on EHR data. To impute data, a general approach will start with a RNN model with LSTM layers [38][39][40][41][42][43], or GRU layers [44]. In more advanced approaches, researchers can modify the architecture parameters based on the dataset size or augment the model with additional pieces.…”

Section: Applicationsmentioning

confidence: 99%

Generative AI models in time varying biomedical data: a systematic review (Preprint)

He,

Sarwal,

Qiu

et al. 2024

Preprint

View full text Add to dashboard Cite

BACKGROUND Trajectory modeling is a longstanding challenge in the application of computational methods to healthcare. However, traditional statistical and machine learning methods do not achieve satisfactory results as they often fail to capture the complex underlying distributions of multi-modal health data, and long-term dependencies throughout patients’ medical histories. Recent advances in generative AI have provided powerful tools to represent complex distributions and patterns with minimal underlying assumptions. These have had a major impact in fields such as finance and environmental sciences, and recently researchers have turned to these methods for disease modeling. OBJECTIVE While AI methods have proven powerful, their application in clinical practice remains limited due to their black-box like nature. The proliferation of AI algorithms poses a significant challenge for non-developers to track and incorporate these advances into clinical research and application. In this work, we survey peer-reviewed, generative AI model papers with specific applications in time series health data. METHODS Our search includes single- and multi-modal generative AI models that operate over structured and unstructured data, medical imaging or multi-omics data. We introduce current generative AI methods, review their applications in each data modality and discuss their strengths and weaknesses compared to traditional methods. RESULTS We follow the PRISMA guideline and review 155 articles on generative AI applications in time series healthcare data across modalities. Furthermore, we offer a systematic framework for clinicians to easily identify suitable AI methods for their data and task at hand. CONCLUSIONS We review and critique existing applications of generative AI to time series health data with the aim of bridging the gap between computational methods and clinical application. We also identify shortcomings of existing approaches, and highlight recent advances in generative AI that represent promising directions for healthcare modeling.

show abstract

“…Although these methods can predict EC numbers for proteins without similar references, the prediction speed and precision are not ideal. Because deep learning has delivered powerful results in many areas [ 12 – 15 ], researchers use deep learning methods to predict EC numbers and continually improve the precision of functional annotation [ 2 ]. However, deep learning methods are prone to overfitting because of an unbalanced distribution of training datasets [ 16 ].…”

Section: Introductionmentioning

confidence: 99%

Enzyme Commission Number Prediction and Benchmarking with Hierarchical Dual-core Multitask Learning Framework

Shi,

Deng,

Yuan

et al. 2023

Research

Self Cite

View full text Add to dashboard Cite

Enzyme commission (EC) numbers, which associate a protein sequence with the biochemical reactions it catalyzes, are essential for the accurate understanding of enzyme functions and cellular metabolism. Many ab initio computational approaches were proposed to predict EC numbers for given input protein sequences. However, the prediction performance (accuracy, recall, and precision), usability, and efficiency of existing methods decreased seriously when dealing with recently discovered proteins, thus still having much room to be improved. Here, we report HDMLF, a hierarchical dual-core multitask learning framework for accurately predicting EC numbers based on novel deep learning techniques. HDMLF is composed of an embedding core and a learning core; the embedding core adopts the latest protein language model for protein sequence embedding, and the learning core conducts the EC number prediction. Specifically, HDMLF is designed on the basis of a gated recurrent unit framework to perform EC number prediction in the multi-objective hierarchy, multitasking manner. Additionally, we introduced an attention layer to optimize the EC prediction and employed a greedy strategy to integrate and fine-tune the final model. Comparative analyses against 4 representative methods demonstrate that HDMLF stably delivers the highest performance, which improves accuracy and F1 score by 60% and 40% over the state of the art, respectively. An additional case study of tyrB predicted to compensate for the loss of aspartate aminotransferase aspC, as reported in a previous experimental study, shows that our model can also be used to uncover the enzyme promiscuity. Finally, we established a web platform, namely, ECRECer ( https://ecrecer.biodesign.ac.cn ), using an entirely could-based serverless architecture and provided an offline bundle to improve usability.

show abstract

Deep dynamic imputation of clinical time series for mortality prediction

Cited by 16 publications

References 35 publications

Deep multi-modal intermediate fusion of clinical record and time series data in mortality prediction

Deep multi-modal intermediate fusion of clinical record and time series data in mortality prediction

Generative AI models in time varying biomedical data: a systematic review (Preprint)

Enzyme Commission Number Prediction and Benchmarking with Hierarchical Dual-core Multitask Learning Framework

Contact Info

Product

Resources

About