Big Data on Healthcare Application

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 7 of 7
  • Item
    Using Explainable Deep Learning and Logistic Regression to Evaluate Complementary and Integrative Health Treatments in Patients with Musculoskeletal Disorders
    ( 2020-01-07) Redd, Douglas ; Goulet, Joseph ; Zeng-Treitler, Qing
    There is an increasing interest in the use of Complementary and Integrative Health (CIH) for treatment of pain as an alternative to opioid medications. We use a novel explainable deep learning approach compared and contrasted to a traditional logistic regression model to explore the impact of musculoskeletal disorder related factors on the use of CIH. The impact scores from the neural network show high correlation with the log-odds ratios of the logistic regression, showing the promise that neural networks can be used to identify high impact factors without depending on a priori assumptions and limitations of traditional statistical models.
  • Item
    Blood Glucose Forecasting using LSTM Variants under the Context of Open Source Artificial Pancreas System
    ( 2020-01-07) Wang, Tianfeng ; Li, Weizi ; Lewis, Dana
    High accuracy of blood glucose prediction over the long term is essential for preventative diabetes management. The emerging closed-loop insulin delivery system such as the artificial pancreas system (APS) provides opportunities for improved glycaemic control for patients with type 1 diabetes. Existing blood glucose studies are proven effective only within 30 minutes but the accuracy deteriorates drastically when the prediction horizon increases to 45 minutes and 60 minutes. Deep learning, especially for long short term memory (LSTM) and its variants have recently been applied in various areas to achieve state-of-the-art results in tasks with complex time series data. In this study, we present deep LSTM based models that are capable of forecasting long term blood glucose levels with improved prediction and clinical accuracy. We evaluate our approach using 20 cases(878,000 glucose values) from Open Source Artificial Pancreas System (OpenAPS). On 30-minutes and 45-minutes prediction, our Stacked-LSTM achieved the best performance with Root-Mean-Square-Error (RMSE) marks 11.96 & 15.81 and Clark-Grid-ZoneA marks 0.887 & 0.784. In terms of 60-minutes prediction, our ConvLSTM has the best performance with RMSE = 19.6 and Clark-Grid-ZoneA=0.714. Our models outperform existing methods in both prediction and clinical accuracy. This research can hopefully support patients with type 1 diabetes to better manage their behavior in a more preventative way and can be used in future real APS context.
  • Item
    High-performance Diagnosis of Sleep Disorders: A Novel, Accurate and Fast Machine Learning Approach Using Electroencephalographic Data
    ( 2020-01-07) Buettner, Ricardo ; Grimmeisen, Annika ; Gotschlich, Anne
    While diagnosing sleep disorders by physicians using electroencephalographic data is protracted and inaccurate, we report promising results from a novel, fast and reliable machine learning approach. Our approach only needs an electroencephalographic recording snippet of 10 minutes instead of eight hours to correctly classify the disorder with an accuracy of over 90 percent. The Rapid Eye Movement sleep behavior disorder can lead to secondary diseases like Parkinson or Dementia. Therefore, it is important to classify the disorder fast and with a high level of accuracy - which is now possible with our approach.
  • Item
    Predicting Risk of Hospital Readmission for Comorbidity Patients through a Novel Deep Learning Framework
    ( 2020-01-07) Dashtban, M ; Li, Weizi
    Hospital readmission is widely recognized as indicator of inpatient quality of care which has significant impact on healthcare cost. Thus, early recognition of readmission risk has been of growing interest in various hospitals. Additionally, there has been growing attention to provide better care to patients with more complications, whose care would impact the quality of care in multiple directions. To this regard, this research specifically targets comorbidity patients i.e., the patients with chronic disease. This research proposes a novel deep learning- framework termed SDAE-GAN. The presented approach consists of three phases. Firstly, various groups of variables from heterogeneous sources are collated. These variables mainly include demographic, socioeconomic, some statistics about patient’s frequent admissions and their diagnosis codes. Then, more processing applies dealing missing values, digitization and data balancing. Afterwards, stacked denoising auto-encoders function to learn underlying representation; and technically to forms a latent space. The latent variables then are used by a Generative Adversarial Neural Networks to evaluate the risk of 30- day readmission. The model is fine-tuned and being compared with state-of-the-arts. Experimental results exhibit competitive performance with higher sensitivity.
  • Item
    Configurational Approach to Identify Concept Networks in selected Clinical Safety Incident Classes
    ( 2020-01-07) Gupta, Jaiprakash ; Poon, Simon
    Classifying clinical safety incidents (CSI) in their correct classes depends on the multiple concepts used to describe them. Machine learning based classification case study presented in this paper shows that it fails to identify the underlying complex concepts associations between the CSI classes. Two pairs of classes, each having high and low confused classes (as determined by the classifier), were further investigated by applying the set-theoretic-based logical synthesis methodology. The aim is to identify the relationships between concept networks for selected classes. The concept networks were identified using a set of 117 terms and measures taken included degree-centrality and in-betweenness centrality. In this study, using deterministic configurational approach, it is feasible to draw a meaningful relationship between concepts using the complex medical dataset sourced from the Incident Information Management System. The study is proof of concept that it is possible to identify concept networks and concept configuration rules for CSI classes.
  • Item
    Development of a Machine Learning Based Algorithm To Accurately Detect Schizophrenia based on One-minute EEG Recordings
    ( 2020-01-07) Buettner, Ricardo ; Beil, David ; Scholtz, Stefanie ; Djemai, Aadel
    While diagnosing schizophrenia by physicians based on patients' history and their overall mental health is inaccurate, we report on promising results using a novel, fast and reliable machine learning approach based on electroencephalography (EEG) recordings. We show that a fine granular division of EEG spectra in combination with the Random Forest classifier allows a distinction to be made between paranoid schizophrenic (ICD-10 F20.0) and non-schizophrenic persons with a very good balanced accuracy of 96.77 percent. We evaluate our approach on EEG data from an open neurological and psychiatric repository containing 499 one-minute recordings of n=28 participants (14 paranoid schizophrenic and 14 healthy controls). Since the fact that neither diagnostic tests nor biomarkers are available yet to diagnose paranoid schizophrenia, our approach paves the way to a quick and reliable diagnosis with a high accuracy. Furthermore, interesting insights about the most predictive subbands were gained by analyzing the electroencephalographic spectrum up to 100 Hz.
  • Item
    Introduction to the Minitrack on Big Data on Healthcare Application
    ( 2020-01-07) Hung, Patrick ; Poon, Simon ; Tsoi, Kelvin