Improved methods for the bedside diagnosis and evaluation of neuromuscular disorders are needed. One technology that is finding increasing use for this purpose is electrical impedance myography (EIM). In EIM, a very weak, high frequency electrical current is passed through a muscle of interest and the resulting surface voltages are measured. Disease associated alterations in the composition and microstructural features of the muscle produce characteristic changes that can be used to help classify specific conditions and grade disease severity. To date, most studies using EIM analysis have utilized a fairly limited data set for disease assessment. While effective, this approach ignores a great deal of information locked within the impedance data, including those values that can assist in predicting specific muscle features (such as myofiber diameter) and the presence of pathological change (e.g., fat or connective tissue deposition). In addition, as it stands, the data set is challenging for the clinician to understand without a detailed knowledge of impedance theory. Myolex, Inc is a small business concern located in Boston, MA has as its main focus the development of EIM technologies for clinical use. Myolex recently completed a Phase 1 SBIR that demonstrated the potential capability of machine learning based classification algorithms to effectively discriminate healthy muscle from diseased and to discriminate one disease from another. In this proposed work, we will greatly advance this concept by embodying classification algorithms into a powerful new software suite for Myolex?s current EIM system, the mView. Our underlying hypothesis is that EIM data analysis can be automated to the point that classification systems can provide data on disease diagnosis as well as disease severity for improved ease-of-use. We propose to study this hypothesis via 2 specific aims.
In Specific Aim 1, we will design a software suite capable of assisting with artifact-free data collection to be incorporated into our current EIM system, the mViewTM. Then using classification paradigms based on a prodigious amount of previous collected data, we will develop an automated data analysis tool to help provide data on disease category as well as microscopic features, muscle based on the impedance data alone using Microsoft?s Azure Cloud platform.
In Specific Aim 2, we will test this developed software suite in a total of180 adult and pediatric neuromuscular disease patients and healthy participants evaluated at Ohio State University Wexner Medical Center (adults) and Boston Children?s Hospital (children). During this data collection period, the Ohio State and Boston Children?s researchers will have real- time access to Myolex staff to provide feedback and have questions/problems answered and addressed. The user interface will continue to be refined and classification algorithms improved. At the conclusion of this work, a new diagnostic tool will be developed for potential 510(k) FDA approval. It will serve as the basis for a continuously self-refining system as additional data sets are collected by end-users employing them in regular clinical use.
Electrical impedance myography (EIM) is a valuable technique to assist with the evaluation of a variety of conditions affecting nerve and muscle. However, to date, only simplistic EIM outcomes have been utilized to assess muscle condition. In this proposed work, we will develop a software platform using machine learning to be incorporated into current EIM technology to allow for automated diseased classification and characterization using the entire large EIM data set collected with each muscle measurement. This will serve as the basis for a new, powerful and convenient tool for neuromuscular diagnosis that will continue to advance over time.
