Circuit Dynamics and Computational Neuroscience I.1.l Deep and machine learning Monday AM + Wednesday AM

Abstract

Abstract Body

Aims: Axonal degeneration (AxD) is a pathological hallmark of neurological diseases. Although different types of AxD have been described, the underlying morphological differences are not completely understood.

Methods: We developed a microfluidic device to study axons in vitro and the EntireAxon neural network to segment axons on phase-contrast microscopic images and to quantify morphological features of AxD including axonal swellings and axonal fragments in a high-throughput manner.

Results: The EntireAxon sensitively and specifically detected all features of AxD and its performance exceeded that of human expert ratings. In hemin-induced AxD, we detected a concentration- and time-dependent degeneration of axons. The axon area was significantly smaller starting at 10.5 (200 μM; p<0.001), 11.5 (100 µM; p<0.001) and 14 hours (50 µM; p=0.001) compared to control conditions (0 µM). Fragment area and AxD index (ratio of the fragment area over total axon area including swellings and fragments) were significantly increased starting at 6 hours after hemin exposure (200 µM vs. 0 µM). Axonal swelling area was, however, not significantly different between the concentrations used. Using a recurrent neural network, we further identified four morphological types of AxD: granular, retraction, swelling, and transport degeneration.

Conclusions: We show that axons exposed to the same hemin concentration undergo concomitantly multiple, morphologically different types of AxD. This approach reveals the morphological heterogeneity of AxD in neurological diseases. Our approach will allow the systematic analysis of AxD and unravel the intricacy in which AxD occurs in hemorrhagic stroke and potentially other neurological diseases.

Hide