NCCID Project Summaries

The core of the NCCID initiative is to provide value to the public in response to the COVID-19 crisis. For this reason, NCCID wishes to share information on how the data is being utilised by approved institutions and researcherts, to inform the wider community of patients, staff and interested public.

Below is a list of the projects currently ongoing.

University College London

University College London plans to store the NCCID data in a highly secure XNAT repository to enable imaging-based research at UCL Centre for Medical Image Computing (CMIC). In particular, researchers at CMIC will use these data to build artificial intelligence models to automatically detect COVID-19 patients based on their CT or X-ray images, such that, in the future, these images can be screened automatically before doctors read them. This will significantly save time in managing future outbreaks. The research also involves building computational models to analyse the outcomes for those with confirmed COVID-19 diagnosis, predicting best management for individual patients. These predictions may shorten their hospital stay, reduce complications and even save lives. Finally, the project will investigates methods to deploy these developed models to local hospitals quickly and safely.

University of Cambridge

It is strongly believed that early detection of COVID-19 and intervention leads to lower Covid-19 mortality because it enables disease treatment via oxygen therapy and control of spread via isolation. The diagnosis of COVID-19 must be confirmed by reverse-transcription polymerase chain reaction (RT-PCR) or gene sequencing for respiratory or blood specimens. However, testing the general population is proving to be very challenging because of various reasons including limitations of sample collection and transportation, kit performance and availability, limitations in capacity, etc.

Chest scans could include x-rays, CT and MRI scans. Chest CT scans are used to examine lung tissues and often used for further investigation after an abnormal chest x-ray. Chest MRI scans provide a detailed picture of the chest wall, heart, and blood vessels. These scans are carried out routinely for a variety of medical reasons, including preparation for surgery, annual follow-ups, accident and emergency, etc. The creation of computer systems that can automatically process these scans to detect and identify signs of Covid-19 can provide added value for the NHS with no significant additional burden on staff, resources, operational costs, etc. The development of these systems require the implementation of cutting-edge image processing and artificial intelligence technologies.

X-ray images and CT scans can also be useful in monitoring the progression of Covid-19 patients as they can reveal if their lungs are filled with sticky mucus that can lead to breathing problems and provide a benchmark for comparisons with previous scans.

This project aims to find the visual signatures of Covid-19, as they appear in chest scans, that can lead to accurate diagnostic and prognosis for use in hospital settings. Automated imaging algorithms, aided by advanced artificial intelligence techniques, can detect some of the abnormal features appearing in these scans, such as ground glass patterned areas, Ground glass, Crazy paving, Vascular dilatation, Traction Bronchiectasis. These features are generally not specific to Covid-19 and could be seen with other infections. Hence, it is important to develop AI techniques to aid the imaging analysis to increase the accuracy of diagnostic.

University of Bradford

Coronavirus Disease 2019 (COVID-19) is highly contagious, and severe cases can lead to pneumonia and ultimately death. The diagnosis can be confirmed by laboratory testing; however, the test has low sensitivity which leads to late diagnosis and treatment. Chest X-rays and CT scans provide valuable diagnostic and monitoring information that can complement the laboratory and clinical data. In this project, we propose to develop an open-source artificial intelligence tool that combines chest imaging data and clinical data to support the diagnosis, triaging and prognosis for COVID-19 in the UK. This will make clinical decisions more efficient, accurate, timely, and potentially cheaper, leading to better patient outcomes.


Chest CT scans are used in hospitals across the globe to image the severity of COVID-19 lung involvement and guide the appropriate patient management. Artificial intelligence (AI) designed for radiologists can increase the speed of reporting on these scans and support timely patient triaging.

Aidence has set-up an international consortium, ICOVAI, to create an AI solution for COVID-19 on chest CTs. The consortium is a collaboration between clinical centers, hospitals, AI companies, and distribution partners.

ICOVAI's AI solution will automatically detect COVID-19 on chest CTs and assess the extent of lung tissue affected. Its quantitative analysis can be used to guide hospital management, such as bed capacity on wards, or predicting the need for ICU care.

The consortium aims to reduce the workload and pressure that the medical staff are facing during the pandemic. The software will be particularly useful when test kits are absent or inconclusive, and when radiologists are unavailable or lack specific COVID-19 training.

To train a well-performing model, ICOVAI is using high-quality datasets from diverse CT scanners, hospitals, and countries. The patient data is anonymised and processed in line with the GDPR. The product will comply with the Medical Device Regulation (MDR), 2017/745, to ensure clinical safety and quality.

The AI solution will be made available not-for-profit for the NHS and European hospitals. The project is backed by the EU.

City Data Science Institute

The City Data Science Institute are using the NCCID dataset to develop artificial intelligence systems that offer explainable decision making. More specifically, we are investigating the key radiological findings of Covid-19, how this can change over time, and how this differs from other disease findings upon a chest X-ray. Using state-of-the-art generative networks, we aim to learn more about the Covid-19 disease process and facilitate medical decision making.

The chest X-ray is a readily available investigation and is useful in the identification, severity assessment and monitoring of Covid-19. It is more readily available than CT scanning and able to exclude other important conditions that may present. The use of AI to assess chest X-rays can facilitate medical staff and improve patient outcomes.

We have begun using a particular type of artificial intelligence model that is able to learn what an X-ray should appear like if it were to be healthy or more diseased. This will help the research community learn about subtle disease features and can contribute to more accurate and quicker automatic diagnostics. One of our main goals is for our system to offer a counterfactual explanation as to why the artificial intelligence has made certain decisions. This is vital in developing safe and effective AI.

Medical Analytica Ltd

To test, evaluate and further finetune a software which can automatically analyse Chest X-ray images to identify absence or presence of features of COVID-19 infection. The software can detect other lung conditions such as pneumonia caused by viral or bacterial infections. The software is being further developed to identify other key conditions such as lesion and enlarged heart. The software utilises a number of mathematical models for image analysis and is capable of offering a high confidence classification with minimum false positives or false negatives.

Ultimate objective is integrating it into the NHS radiology reporting workflow to provide a prompt computer aided prediction alongside the radiologist's own report. By providing an extra layer of support to the clinical team, patient triage and access to appropriate treatment can be speed up. Saving time, resources and most importantly, improving patient experience.

Other application of the software is to provide preliminary / indicative prediction to the primary care team in remote locations and community hospitals to assist in identifying in the community cases which may need urgent specialist attention in main hospitals.

Universities of Brighton, Oxford, Glasgow, Lincoln and Sheffield

Members of this collaboration were instrumental in winning first place in 'Coronahackathon' April 2020, for the development of Machine Learning (ML) and Artificial Intelligence (AI) to predict patients with SARS-CoV-2 virus using full blood count results. SARS-CoV-2 positive patients exhibit a characteristic change in different parameters measured in simple and rapid blood tests to a high accuracy, predicting the virus in regular wards (93-94%) and those in the community (80-86%).

Our project will validate these initial results and enable use in current hospital practice to screen patients and identify those needing full diagnosis for SARS-CoV-2. Expertise in chest images, blood science and modelling ML and AI will develop an innovative tool to upscale screening to identify individuals for full rt-PCR testing of the virus potentially up to one week earlier than rt-PCR, which will allow much faster release of the country (and the world) from lockdown, protection against future waves and future pandemics.

Ashford and St Peter's Hospitals NHS Foundation Trust

Chest X-rays are often one of the first tests used to help guide doctors with deciding how likely a patient is to have COVID-19 and also how severe the infection is. We are aiming to see how accurate chest X-rays are for detecting COVID-19 and telling doctors how severe the disease is. We plan to answer this by having doctors specialising in X-rays assess chest X-rays of patients in the national COVID-19 imaging database and then compare this assessment to clinical details, such as if they had COVID-19 and how well they did in hospital. We hope this will give us a better understanding of chest X-ray accuracy in COVID-19 and appearances that are linked to COVID-19 or more severe infection. We also additionally aim to use Neural networks (advanced computational algorithms) on the chest X-rays in the database to see if these can be used to automatically detect COVID-19 in chest X-rays. This research will hopefully one day help with the development of clinical algorithms and technology that can be used to speed up chest X-ray assessment for COVID-19. is currently operating existing services within the NHS for the triage of chest X-rays based on the presence of abnormalities, as determined by artificial intelligence algorithms. In this study aims to achieve two key goals.

First, we aim to validate that an algorithm developed for the general detection of abnormalities on plain-film chest X-rays can identify novel forms of abnormality (such as COVID-19) despite being developed prior to that pathology's initial presentation (in this case, prior to February 2020). We propose that this will show that the use of existing AI-based patient triage can assist healthcare systems in the efficient use of resource and radiological capacity during high stress periods such as pandemic flu, even when novel pathologies are present.

Secondly, in recognition of the increased utilisation of AI across healthcare sectors, we aim to establish a set of best practices for rapidly and regularly updating clinically deployed algorithms in order to enable efficient 'learning' of the characteristics of novel pathologies. This will be validated by the ability of existing chest X-ray models to learn the characteristics of COVID-19 as demonstrated in the NCCID. We suggest that this work will ensure that the implementation of AI into health systems improves future pandemic preparedness as well as maximising responsiveness to population health.

The Royal Surrey NHS Foundation Trust

Chest imaging is increasingly used as an alternative method for screening COVID-19, with high sensitivity compared to laboratory testing methods. AI tools have the potential to enable fast and accurate diagnosis from chest X-rays (CXR). However, several issues of image quality have been identified as a limitation to the diagnostic performance of CXR, for example, in images acquired on portable machines, and where under-exposure occurs due to patient positioning or patient BMI. Little is known about the impact of image-quality upon the accuracy and sensitivity of AI algorithms, and we propose to investigate this, having been significantly involved in the development of the NCCID database, and having successfully led previous evaluations of AI tools, for example for diagnosis in breast screening.

The aim of this study is to evaluate the impact of image quality on the performance of AI models by analysing performance metrics during training and validation using the NCCID dataset. Published, open-source models shown to classify COVID-19 on CXR will be identified and the best-performing algorithm selected by assessing performance on the NCCID test dataset. The NCCID training and test datasets will be analysed using existing image analysis tools developed in-house at RSNFT, enabling the categorisation of the NCCID dataset according to image quality, and the dataset will be stratified into datasets of different image quality. The AI model will be retrained and tested on different combinations of training and test data image-quality and the impact of image-quality upon the accuracy and sensitivity of the model evaluated.

Imperial College London

Sarcopenia is defined as the loss of skeletal muscle mass or function, is primarily a disease of the elderly and a marker of frailty. Currently, physical fitness is assessed with performance status, however this score is subject to interobserver variability. There is increasing interest in the clinical importance of sarcopenia in a wide range of malignancies such as lung, breast, upper GI and colorectal cancers, however the relationship to patient outcome in COVID-19 has not been evaluated. CT provides an objective and easily reproducible assessment of skeletal muscle which has been validated in cancer patients. These studies used a time consuming manual segmentation of skeletal muscle at L3 (3rd lumbar vertebra level) which is not easily integrated into patient care and radiology reports. The research team have already developed a fully automated technique for measuring sarcopenia on CT at L3. The aim of the study is to evaluate an objective fully automated assessment of muscle area and adiposity using artificial intelligence (AI) deep learning techniques on CT to determine whether these measures are linked to patient outcome in COVID-19 patients.

University of Greenwich - School of Computing and Mathematical Sciences

The project deals with diagnosis of COVID-19 from chest CT scans, scan series and x-rays, through a novel Deep Learning methodology. Its main target is to support unification of the rather fragmented field in UK and internationally, where projects train DL systems over specific datasets, obtained in smaller local, or larger national frameworks, with no proof of good generalisation over different datasets and different clinical environments. It will achieve this through generation of a portable framework for COVID-19 diagnosis, based on analysis of information extracted from trained deep neural networks and adaptation across different data cohorts and input modalities.

In this framework, the main systems that have been trained with chest imaging examinations for diagnosis of COVID-19 in Greece will be tested and adapted to the UK cohort of CT scans and x-rays. It will be examined whether the systems trained in a European country (such as Greece) perform also well when applied to data from UK patients. Either, or both CT scans and x-rays will be provided as inputs to these systems, for single modal, or multi-modal COVID-19 detection. Furthermore, it will be investigated whether this performance gets improved, if the systems get adapted to the UK cohort. Finally, a unification step will be implemented, by merging both systems, i.e.,the original and the adapted ones, also validating the good performance of the unified system on the UK data Cohort. This will illustrate the portability of the developed unified model across Greece and UK and will serve as pilot that can be extended with more COVID-19 diagnosis models from other parts of Europe, or elsewhere.


As healthcare systems around the world face unprecedented stress levels, support to the clinical teams responsible for diagnosis and treatment of COVID19 patients is a priority. Recently, Philips (Royal Philips N.V.) has developed an automatic AI-based processing application (called “CT Pulmo Auto Results”, released in the USA under FDA Emergency Use Authorization and under regulatory review in EU) to assist clinical teams in quantification of the level of COVID-19 pneumonia disease burden. With the objective to further improve and monitor the performance of the application as the current pandemic unfolds, Philips is currently testing its application in collaboration with healthcare stakeholders at multiple locations around the world. As AI-based algorithms are claimed to be dependent on the training cohort Philips looks seriously to evaluate on various different cohorts. Within that scope, the National COVID-19 Chest Imaging Database (NCCID) will be used to study the performance of the application in UK patient datasets. Although approved for use in UK, the regulatory approval CE marking was not based on a UK population. In light of regulatory changes with CE marking required in future submissions and new MHRA policies coming into effect we also wish to learn from the NHSx AI lab on how to work effectively together on future R&D studies in the UK using this study as an example and to work out any challenges together. Furthermore, we would explore the combination of the CT-based imaging volumetric measurements with the collected non-imaging patient data for COVID-19 pneumonia characterization with the goal to explore correlations with clinical endpoints, such as admission to Intensive Care Units or duration of stay in nursing wards.


A crucial element in controlling the spread of COVID-19 is effective screening and timely treatment. Early detection of COVID-19 allows for swift intervention with isolation and appropriate use of personal protective equipment improving NHS resource allocation. Current practice commonly relies on PCR testing for confirmation of the diagnosis which takes over 24 hours for results, which can have a sensitivity as low as 60-70%. A supplementary automated tool for early detection of COVID-19 patients based on radiology could be a crucial part of addressing these issues. A wide variety of deep learning models have already been developed to detect COVID-19 in chest X-ray. However, due to a previous scarcity of COVID-19 data, many of the models the models that have been published are at high risk of bias and lack sufficient evaluation. This project will rigorously evaluate existing deep learning models with the aim of identifying the approaches most suited to detecting COVID-19 in chest X-rays. In-depth appraisal of model performances will highlight the pitfalls associated with the problem domain and help guide future work.


Claritas iRAD is an image enhancement software product that provides image enhancement across the modalities of X-ray, CT and MRI. Chest computed tomography (CT) is considered to be the primary diagnostic modality for examining patients with COVID-19, exposing patients to ionising radiation. Image enhancement technology may improve the utility of X-ray and MRI by helping to identify subtle and obscured features, hence reducing the exposure and risks for patients. In addition, iRAD can improve CT image quality, increasing clinically relevant details and reducing the need for repeat scanning. This research will help understand how iRAD can be leveraged to improve diagnosis and image interpretation of COVID -19 and other disease areas such as lung disease and malignancy. This research will also look at how the evolution of a disease phenotype over time may be improved.

All content is available under the Open Government Licence v3.0, except where otherwise stated.