�鶹��Ƶ

A new method based on machine learning for diagnosis of cochlear synaptopathy

Jianxin Bao, Ph.D., Anatomy and Neurobiology, NEOMED

Tsung-Heng Tsai, Ph.D., Mathematical Sciences, �鶹��Ƶ

Identification of functional biomarkers for synaptic loss with noninvasive methods is critical for developing therapies to prevent and treat neurodegenerative diseases such as Alzheimer’s disease. In the auditory system, animal studies have shown that sound-evoked electrophysiological potentials can be used to detect cochlear synaptopathy, an early stage of neurodegeneration which can accelerate age-related neuronal loss. Because cochlear synaptic loss can occur without any measurable changes in audiometric thresholds, this synaptopathy remains hidden from current clinical diagnostics, which leads to a popular term used in the field: “hidden hearing loss.” Based on our preliminary data, here we will obtain critical proof-of-concept data by identifying functional features in mouse models with noise-induced hearing loss (NIHL) in which cochlear synaptopathy can be directly observed through histological quantification. First, we will first determine cochlear synaptopathy in mouse models. The main advantage of NIHL models is that the noise level can be precisely presented to control the occurrence of cochlear synaptic loss. To directly associate identified electrophysiological features with noise-induced synaptic loss, we will quantify the synapse number between inner hair cells and spiral ganglion neurons with or without noise exposure. These data will enable us to define individual mice with cochlear synaptopathy and directly associate these individuals with multiple electrophysiological features. Second, we will identify multi-metrics for cochlear synaptopathy. In our preliminary study, we obtained 7 electrophysiological features highly associated with cochlear synaptopathy. We will expand our functional features by adding a phase-locking value, a metric of neuronal synchrony, to determine if we can further improve our detection sensitivity by non-invasive means.
In short, we will develop a new multi-metric method to detect cochlear synaptopathy by two steps: (1) Identifying electrophysiological features predictive of cochlear synaptopathy in two mouse NIHL models in which cochlear synaptic loss can be directly validated by histological quantification, and (2) Identifying functional features predictive of cochlear synaptopathy using machine learning. If successful, the outcomes of this project could lead to the development of a new clinical diagnostic tool for the detection of hidden hearing loss.