400 million people across the world have chronic obstructive pulmonary disease, also known as COPD- which causes breathing issues consisting of shortness of breath, wheezing and frequent chest infections. Whilst in developed countries, people with lung disease have most probably contracted it through smoking, those in developing countries may contract these diseases through increased levels of pollution, through outdoor fumes or by burning biomass indoors. Another predominant lung condition is asthma, which affects around 300 million people globally. Given the scale of these diseases, many pulmonologists and researchers are investigating whether artificial intelligence has the potential to diagnose these lung conditions early on, to reduce the severity for patients whilst alleviating the backlog for healthcare services.

​

A common device used to diagnose lung conditions is the spirometer, which measures the capacity of one’s lung function. Whilst spirometers have proven to be accurate, they are not universally accessible and usually requires the presence of a healthcare professional to interpret the test.

​

A University College London lecture by Professor John Hurst and Dr Luke Hale delved deeper into the possible alternatives to using spirometers. Professor Hurst discussed the possibility of artificial intelligence in finding a new solution to diagnosing lung conditions. The approach involved using smartphones and modelling them as diagnostic devices with the use of neural networks to predict airflow data from the phone’s audio and camera data.

​

To gather the input data, a person inhales and exhales which is captured through their phone’s microphone and camera. This data is then processed through neural networks, which essentially take the input data and produce an output through the use of nodes, which perform calculations repeatedly to process the data. Each node is connected by weights which detect the strength of the signal between each node and process the data accordingly, ultimately producing the output airflow data. 

​

An app called Lungy has been developed which uses this model designed to educate more people about lung health. The platform can be used to incentivise breathing exercises whilst also allowing people to understand more about their own lungs as they receive real time feedback on their breathing. Although this app can only be used for healthy lungs currently, it is a promising development which may be able to help patients manage their conditions through teaching exercises and encouraging healthier habits. Nevertheless, despite AI’s potential, one must recognise the challenges that come with it. For example, the accuracy of the AI’s neural networks relies on the quality of the input data, meaning poor user technique may lead to inaccurate results.

​

Neural networks are incredibly important in AI, as they are fundamental in the machine learning aspect which focuses on developing algorithms where one can train computers to recognise and predict data accurately. The Lancet highlighted a study published in 2018 which found AI algorithms were just as effective in identifying fibrotic lung disease as thoracic radiologists. 

One key aspect of medicine that can be assisted by AI is diagnostic imaging, with experts like Simon Walsh from Imperial College London stating ‘Images are essentially numerical data and computers are particularly well suited to analysing that kind of data’. As computers are able to detect abnormalities in tissues and predict disease progression with speed and accuracy, there is potential for AI to play a transformative role in medical imaging.

​

References:

​https://www.youtube.com/watch?v=KuHXO60t4o4&embeds_referring_euri=https%3A%2F%2Fwww.ucl.ac.uk%2F&embeds_referring_origin=https%3A%2F%2Fwww.ucl.ac.uk&source_ve_path=MjM4NTE

https://www.thelancet.com/journals/lanres/article/PIIS2213-2600(19)30331-5/fulltext