Investigation of deep learning neural network architectures for biomedical pattern classification problems

点击翻译按钮获取中文摘要

Recently, there have been many amazing pattern classification breakthroughs achieved by deep learning neural networks. These advancements have been made possible due to Graphics Processing Units (GPU) hardware, which significantly accelerates the deep learning algorithms, and by the availability of data in labelled "big data'" datasets. Consequently, more training can occur due to the increased computational power afforded by GPUs, and these large-scale datasets allow for improved partitioning of high dimensional feature space for classification tasks. The result is that trained neural networks can achieve very low pattern classification error rates. Thus, the aim of this Engage Grant is to investigate deep learning neural network architectures for use by Sightline Innovation Inc. to improve and accelerate the performance of its Sightline Perception Engine, which forms the foundation of Sightline's VuPoint biomedical classification products. This research is important as it represents a significant step toward bridging the gap between the growing mountain of personal health information and automated solutions for tailoring an individual's treatment plan. This work is of importance to Sightline, as well as the biomedical industry, as there are currently very few within the biomedical commercial space investigating the integration of machine learning into biomedicine, in vitro diagnostics, and biosecurity. The anticipated outcomes are a deep learning pattern classification system that fits within Sightline's overall product development map, which will facilitate their diagnosis and detection goals. The benefits to Canada include developing automated systems for personalized medicine, and reducing the money spent annually due to infectious diseases and pandemic risk.