Open Collections

PalmGrid, an artificial intelligence approach to automate cylinder task detection

University of British Columbia

Stroke is a common cause of permanent disability accompanied by devastating impairments. Motor, sensory and cognitive deficits are common following stroke, yet treatment is limited. Along with histological measures, functional outcome in animal models has provided valuable insight to the biological basis and potential rehabilitation efforts of experimental stroke. Developing and using tests that identify behavioral deficits is essential to expanding the development of translational therapies. Forelimb Asymmetry Task experiments – often called Cylinder Tests – are used to study the impact of ischemic stroke and its subsequent rehabilitation to contralateral limb movements of studying rodents. Through assessments on qualitative and quantitative aspects of vertical exploration to the Cylinder Wall, extent of locomotor asymmetry is evaluated [25-27]. Traditionally wall rearing assessments are evaluated through manual, stop-watch based measurements that require laborious observations. Methods that automate the process were attempted such as the use of hardware-based sensor detections that passively probes of touches on the sensor grid. In its various implementations, the sensor-based methods fail to specify the limb that rears the wall nor depict the ways forelimbs are coordinated during the rearing. Advent of artificial intelligence (AI) algorithms, notably Deep Neural Networks (DNN), helps to extract posture and coordinates of forelimbs [21]. PalmGrid is the first attempt to exploit AI posture extraction algorithms – based on 50 layers depth in ResNet Deep Neural Networks [29] together with posture extraction algorithm DeepLabCut [24] – to automate the assessment process with 70% detection accuracy using robust, open-source software. Further improvements in deep neural networks precisions, such as increasing its depth or incorporating advanced posture extraction algorithms, will further enhance detection precisions. In this way, we will have viable alternatives to conduct cylinder test experiments without suffering from extra cost burdens and complex calibrations.

Text

2019-04-25

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/69959

Neuroscience

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/