Session S72.4

Artificial Neural Network Based ‘Continuous Feedback Loop’ Platform to Support Multicenter Cardiac Clinical Trials

S Jacob, D Bhandare, C Bhandare, R Aravindhakshan*

Wayne State University
Detroit, MI, USA

Multicenter clinical trials are so complex that traditional rigid concepts of data acquisition and analysis algorithms may culminate in poor trial design, incomplete protocols, unidentified variables and absence of feedback modification. This provides no flexibility to accommodate new, useful and informative predictors during the course of the trial. Thus most of the trials will realize the limitations by the end of the study period. This fact motivated the development of a suite of decision support tools utilizing an Artificial neural network based platform. We conceptualize an ANN based platform for supporting large clinical trials based on a ‘continuous feedback loop’ mechanism to justify the addition/modification/deletion of variables or selection and sub selection of other parameters during the entire course of the trial. This will enable to restructure the way of data acquisition, data analysis and will identify novel hidden clinical correlations and findings. Our model is an ANN based algorithm which will run different possible variable combinations and during every iteration it will select the variables that maximize the information gain. This will improve the classifier performance to achieve a high level of prediction and accuracy. Our proposed ANN is based on a multi level perceptron (MLP) model which is a fully-connected, three layer, feed-forward, perceptron neural network. Feature selection technique of our model aims at selecting a feature set that is most discriminative i.e. provides the best prediction of outcome and is least redundant. This is extremely useful in reducing the dimensionality of the data to be processed by the classifier, reducing execution time and improving predictive accuracy. This approach to feature selection involves ranking the inputs according to their relevance to the ANN prediction accuracy. This will be performed during the entire course of the clinical trial and a more targeted data collection process can be tailored during the course of the trial for a better outcome. The novelty of our design is that it will help not only in the analysis, but also will iteratively improve the data collection process throughout the trial period thus significantly improve the efficiency of the trial outcomes and also will improve the cost and time saving.

(Abstract Control Number: 258)