Trial map : A visualization approach for verification of stroke impairment assessment database

Robotic/mechanic devices have become widely used for various medical assessments recently. While using these devices are beneficial in terms of accuracy and objectiveness, validation and consistency problem may occur when combining these data with traditional clinical information. Here we propose a visualization tool that can summarize the experimental data and compare them with the clinical data, in the stroke impairment assessment domain. This visual tool is based on a neural network ensemble that is trained to match the experimental data with Chedoke-McMaster scale, one of the major outcome measure for stroke impairment and recovery assessment. We compare our ensemble model with ten combinations of different classifiers and ensemble schemes, showing that it outperforms competitors. We also demonstrate that our visualization approach is consistent with clinical information, and reliable in a sense that output of our ensemble can be an estimator for the corresponding clinical data when Chedoke-McMaster scores are missing.