Using natural language processing to automate annotation of radiological images

Lead Research Organisation: University of Warwick
Department Name: Mathematics

Abstract

This project falls under the Medical Imaging section's areas of top priority: "Automated extraction and/or integration of existing and additional information from clinical data/images (e.g. via machine learning and/or mathematical science techniques)".
As imaging underpins most medical decisions in healthcare systems, utilisation of radiology has outstripped resources in many nations, including the United Kingdom (UK). Globally, 3.6 billion medical imaging examinations are performed annually and chest radiographs (x-rays) represent the 40% of these images. Coupled with a low number of radiologists per million population (e.g. 49 in the UK), many countries are currently experiencing a workforce crisis with diagnostic delays and errors as imaging continues to increase. Therefore, there is a clear clinical need to transform how radiologists report chest radiographs and computer-based 'Artificial Intelligence' (AI) is a potential solution.
Under a Wellcome Trust project, the group of Giovanni Montana collected more than 2.5 million images with associated radiological reports. They are currently developing an AI triaging and prioritisation system for chest radiographs, ChestAI, which will significantly improve delays to diagnosis by being able to discriminate between normal and abnormal radiographs, and categorise abnormal imaging findings by criticality level to enable automatic prioritisation of critical abnormal radiographs for urgent expert review.
Once the triaging system is validated, the next natural step will be to provide an automatically generated radiology report with all the relevant findings in a human readable manner. With this objective in mind, the group of Giovanni Montana developed an annotation tool for radiological sentences. By the beginning of this PhD they expect to have more than 10K radiological sentences annotated, being the largest dataset of this nature.
This PhD project will focus on the field of Natural Language Generation (NLG). NLG is still an open field, lately being dominated by Neural Networks. This PhD will leverage the state-of-the-art methods on natural image captioning and develop new approaches for the specific field of radiological reports. At the end of this PhD the NLG algorithm will be deployed in several NHS Trusts, currently forming part of the ChestAI project.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S515681/1 01/10/2018 30/09/2022
2159023 Studentship EP/S515681/1 24/09/2018 30/09/2022 Yuanyi Zhu