An Intelligent Assistant for Medical Doctors when Prioritising Pathology Results

在任何一家医院，每年都会产生数百万份电脑化的病理结果。医生检查这些结果，以确定是否需要采取某些行动或仅仅认可结果所需的资源是大量的。这一问题由于检测结果虚假或潜在疾病的风险水平不确定等其他问题而变得更加复杂。因此，优先考虑病理结果是一项重大挑战。该博士研究项目旨在利用机器学习的工具和技术来帮助确定病理结果的优先级，从而节省医生的时间。更具体地说，这项工作是针对使用从病理历史记录中提取的模式来优先考虑以前未见过的病理结果。一个特别的挑战是，可用的数据是未标记的；因此，机器学习必须以无监督的方式进行，或者使用代理来获取基本事实。一种想法是通过识别异常值来确定优先级，另一种想法是识别现有患者记录中导致护理变化的模式和趋势，并将这些模式和趋势用作代理机制。

In any hospital millions of computerised pathology results are produced every year. The resource required by doctors to check these results, so as to determine whether some action is required or simply to endorse the result, is substantial. The issue is compounded by additional issues such as spurious test results or uncertain risk level for potential diseases. Prioritising pathology results is thus a significant challenge. This PhD research project is directed at the utilisation of the tools and techniques of Machine Learning to help prioritise pathology results to save doctor time. More specifically the work is directed at using patterns extracted from pathology historical records to prioritise previously unseen pathology results. A particular challenge is that the data available is unlabelled; the machine learning must therefore be conducted in an unsupervised manner or using a proxy for a ground truth. One idea is to priortise through the identification of outliers, another is to identify patterns and trends in existing patient records that have led to changes in care and use these patterns and trends as a proxy mechanism.