Abstract Details

Title Machine-learning Techniques Applied to F-wave Waveforms Can Predict the Likelihood and Survival of ALS Patients

Topic Neuromuscular and Clinical Neurophysiology (EMG)

Presentation(s) P6 - Poster Session 6 (8:00 AM-9:00 AM)

Poster/Presentation
Number 11-003

Objective To apply machine learning techniques on nerve conduction F-Wave waveforms to predict likelihood and survival of ALS.

Background F-Wave responses can provide critical information for identifying ALS. Machine learning techniques can utilize the time-series F-Wave waveforms to predict the likelihood a patient has ALS. These techniques could enable novel tools to assist clinicians with managing the clinical course of disease.

Design/Methods F-Wave waveforms were analyzed from 47,600 patients who had electrodiagnostic testing (5,310 had motor neuron disease (MND)). Waveforms from ulnar, median, fibular, and tibial F-Wave sessions were processed using wavelet techniques. A Gradient Boosting Machine was trained on statistics from wavelet coefficients, age, sex, and BMI to predict MND. The model was tested against a dataset of 686 confirmed ALS patients meeting Gold Coast Criteria and age/sex matched controls. Cox proportional hazards (CPH) survival analysis was performed to determine significant risk factors for survival.

Results

For the training set, the median age was 64.8 years and 36.5% female for MND; for the test set, the median age was 51.8 years and 56.3% female. Model accuracy was 88% with 87% precision at predicting ALS in the test set. The wavelet model performed better than a cursor-based model that we have previously reported. The model’s log-odds ratio was a significant risk factor in the CPH model (p<0.001) along with age at onset (p<0.001), family history of ALS (p<0.05), and family history of ALS+FTD (p<0.005). Model scores two standard deviations below the median had 0.8 years increased survival (two standard deviations above the median had 0.6 years decreased survival).

Conclusions F-Wave waveforms contain rich information that machine learning techniques can use to predict the likelihood and survival of ALS. This electrodiagnostic information can be used to develop automated tools to help diagnose and manage treatment for ALS patients.

Authors/Disclosures
Kevin A. Mazurek, PhD (Mayo Clinic) PRESENTER	Dr. Mazurek has received personal compensation for serving as an employee of Delos Living / Well Living Lab. Dr. Mazurek has received intellectual property interests from a discovery or technology relating to health care. An immediate family member of Dr. Mazurek has a non-compensated relationship as a Editor with AAN Resident and Fellows Section that is relevant to AAN interests or activities.
Jennifer M. Martinez-Thompson, MD, FAAN (Mayo Clinic)	Dr. Martinez-Thompson has nothing to disclose.
Carolina Parra Cantu, MD (Washington University in St. Louis)	Dr. Cantu has nothing to disclose.
Venkatsampath Raja Gogineni (Mayo Clinic)	Venkatsampath Raja Gogineni has nothing to disclose.
Hugo Botha, MD (Mayo School of Graduate Medical ��ɫ��, Rochester)	Dr. Botha has received research support from NIH. An immediate family member of Dr. Botha has received personal compensation in the range of $500-$4,999 for serving as a Study Section Member with NIH.
David T. Jones, MD (Mayo Clinic)	Dr. Jones has stock in Cephlodyne Neurotechnologies, Inc.. Dr. Jones has received intellectual property interests from a discovery or technology relating to health care.
Ruple S. Laughlin, MD, FAAN (Mayo Clinic Rochester)	Dr. Laughlin has nothing to disclose.
Leland Barnard (Mayo Clinic)	Leland Barnard has nothing to disclose.
Nathan P. Staff, MD, PhD, FAAN (Mayo Clinic)	Dr. Staff has received personal compensation in the range of $500-$4,999 for serving as an Editor, Associate Editor, or Editorial Advisory Board Member for Stem Cell Research & Therapy. Dr. Staff has received research support from National Institutes of Health.

��ɫ����

��ɫ����

��ɫ��

��ɫ��