Deep Metric Learning with Triplet Networks: Application to Hand-grip Myotonia Quantification

Abstract

Myotonia, which refers to delayed muscle relaxation after contraction, is the main symptom of myotonic dystrophy patients. The relaxation time after a hand squeeze has been used as a biomarker for diagnostic purposes and in clinical trials to quantify the effectiveness of a treatment. Current processes that rely on handcrafted features tend to be sensitive to data acquisition noise and intra- and inter-patient variability. In this work, we develop a deep metric learning framework for analyzing the hand-grip time series based on triplet-networks. Experiments show that the learned embedding space can be used to quantify the symptoms, evaluate the effectiveness of treatments, and design new data collection protocols.