CURRENTLY UNDER CONSTRUCTION

Overview

The following documentation provides the protocol and necessary files to apply two hybrid predictive models, which use GUIDE forest, to classify medical imaging studies into one of six categories: flexion, neutral-flexion, neutral, neutral-extension, extension, or extension-flexion based on 17 head position measures.

Background

The Vocal Tract Development Lab at the Waisman Center developed this protocol for the purpose of classifying medical imaging studies (CT/MRI). Confidently classifying head position is vital to conducting analysis of the head and neck. Through the development of this protocol the impact of the dynamic and potential dichonomy of head and neck mobility was considered and the multiple measure approach was found to be more reliabile in classifying head position.

The following protocol utilizes 17 measurements of the head and neck. Additionally, this protocol entails two models that each have two steps. The fist step identifies most of the extension cases, while the second step classifies the remaining imaging studies as flexion, neutral, or extension. Each hybrid model was developed to approach neutral differently; model one focuses on neutral precision while model two focuses on neutral sensitivity. When the two models do not agree it is possible to identify boarderline cases, ex. neutral-flexed or neutral-extension.

General

The training dataset for each step that includes GUIDE forest consists of MRI scans from six participants in five different head positions; extension, flexion, neutral with open mouth, neutral with closed mouth, and neutral with standardized open mouth. All scans were collected prospectively at University of Wisconsin Hospital and Clinics.

The development and validation of this head position classification protocol has been submitted for publication. Once the manuscript is published the citation will be listed here.