COVID-19 detection from chest X-rays using deep learning
This project is based at Neoventive Solutions in Pakistan
- Project title: COVID-19 detection from chest X-rays using deep learning
- Based at: Neoventive Solutions
- Start date: June 2020
- End date: August 2021
- Principal investigator: Dr Tabish Hazir
- Project team: Dr Hana Mahmood, Nabit Bajwa, Hira Kianai, Syed Yahya Sheraz, Syed Ahmar Shah, Suleman Khan
A quick and accurate diagnosis is essential during a pandemic, such as COVID-19. It leads to better outcomes for patients, and can relieve pressure on health care systems struggling to deal with an increasing rate of infection.
The current preferred method for diagnosis of COVID-19 is polymerase chain reaction (PCR). However, some of the hardest hit areas are unable to source enough kits to meet demand and many countries are unable to process tests due to inadequate lab facilities.
Deep learning models, a form of artificial intelligence (AI), are being widely researched and adopted for detection and diagnosis across a variety of diseases.
In this instance, deep learning techniques could be used to identify infected patients using chest X-ray images – which are widely available worldwide. This method could be used in areas where the PCR diagnostic method is not currently feasible.
The use of deep learning to analyse the X-rays could greatly reduce the length of time taken to diagnose patients – with an AI model processing up to 200 images in the average time taken for a radiologist to analyse one.
Aim and impact
The aim of this study is to develop and test a reliable diagnostic tool, using deep learning technology to detect COVID-19 features from chest X-rays. This tool would accelerate the diagnosis and referral of patients, improving clinical outcomes.
Releasing the deep learning model as open source would facilitate the use of the tool both now and in any future pandemics, where a similar algorithm could be used.
This particular application also has the potential to be scaled up and used for more generalised high-impact applications in biomedical imaging.
- Testing dataset has been constructed.
- Model has been validated on both internal and external datasets.