Infrared (IR) spectroscopic imaging and digital pathology for development of algorithm for early detection of test-article related lung findings in th

我们是否可以应用深度学习（AI）来训练一种算法来识别治疗和对照大鼠肺组织红外光谱图像中的异常区域？我们能否开发一种标准化的算法，可以应用于从任何对照和治疗大鼠组织中提取的红外光谱图像，以便在组织病理学病变明显之前识别不良结果。我们能否利用红外光谱图像的数据和经过训练的算法来识别24小时吸入研究中大鼠肺部的不良呼吸结果。使用开发的算法更早地发现治疗后肺组织的不良反应，是否会使先正达减少动物研究或在更早阶段终止不成功的研究？该项目将继续之前的博士项目的工作，开发一种算法，用于检测从最近28天农用化学品吸入研究中提取的对照和处理过的大鼠/小鼠肺组织中的不良结果。虽然组织组成和成分会有所不同，但在小鼠肝脏研究中开发的分析和建模技术很容易转移到肺部，并且样品制备和数据收集协议保持不变。如果这些技术可以被证明对不同的组织组和病变类型有效，它可能会导致标准化化学病变检测算法的发展，并在未来的病理工作流程中实施。这主要是一个计算和人工智能项目，旨在建立在之前博士生的工作基础上。该项目将探索新的方法和新的网络架构，使我们能够从每个样本中提取最大数量的信息。新的科学将是在算法的发展。这在EPSRC的职权范围内，预计学生将具有计算科学/人工智能背景

Research questionsCan we apply deep learning (AI) to train an algorithm to recognize areas of abnormality in treated and control rat lung tissue infrared spectroscopic images?Can we develop a standardized algorithm that can be applied to infrared spectroscopic images taken from any control and treated rat tissue to enable recognition of adverse findings before histopathology lesions are apparent.Can we use the data from the infrared spectroscopic images and the algorithm trained to recognize adverse respiratory findings in lungs taken from rats on 24 hour inhalation studies. Will the earlier detection of adverse findings in treated lung tissue using the developed algorithm enable Syngenta to conduct fewer animal studies or terminate unsuccessful studies at an earlier stage?This project will continue work from a previous PhD project to develop an algorithm for the detection of adverse findings within control and treated lung tissue taken from recent 28 day inhalation studies with agrochemicals, in rats/mice. While the tissue makeup and constituents will differ, the analytical and modelling techniques developed in the mouse liver study are easily transferrable to lung and the sample preparation and data collection protocols remain unchanged. If these techniques can be proven to be effective across different tissue groups and lesion types, it could lead to the development of a standardised chemical lesion detection algorithm to be implemented in future pathology workflows. This is largely a computational and AI project which aims to build on work from a previous PhD student. This project will explore new methods and new network architectures that will enable us to extract the maximum amount of information from each sample. The novel science will be in the algorithm development. This is within the EPSRC remit and it is expected that the student will have a computational science / AI background