AI in detecting lung lesion

After we had looked through Pneumothorax, we will take a look at lung fibrosis. Most lung fibrosis is from interstitial lung disease. Dr. Lee will first explain interstitial lung diseases such as fibrosis. Next, he will explain how AI is utilized in diagnosis.

Doctors rely on high-resolution CT patterns, to identify which disease is lung fibrosis. Hence, many research teams are developing deep learning to classify fibroid lung disease on high-resolution CT.

This is a typical presentation of systemic fibrosis of the lung. You could see the tree of diagnoses on the slide, which covers a whole range of interstitial lung diseases (ILD). Please check this website for more information on the website, ILD.

Other common situations of lung fibrosis happened when people get infections, such as Covid-19 infection or pneumonitis. Some of the patients will develop ARDS, acute respiratory destroy syndrome. This condition will induce lung fibrosis. When the patient got recovered from Covid-19 pneumonia, and also some other diseases, the most famous is idiopathic pulmonary fibrosis(IPF), or we called IPF. IPF is a disease that we don’t know the etiology, which means we don’t know the cause of the disease. However, when the patient comes to the doctors, the presentation shows lung fibrosis.

At a very early diagnosis process, doctors rely on high-resolution CT patterns, to identify which disease is lung fibrosis. AI on image study can help clinical doctors. Dr. Lee will explain the research data published in Lancet respiratory medicines, from Keynes College, London. They developed deep learning to classify fibroid lung disease on high-resolution CT. In this study, they cover a huge number of our patient’s images, and they separate them into three different groups for testing for validation. So this is how the software does, they pick up 4 different levels of the lung, from the upper, middle, lower middle, and the lower level of the lung, to get a full segment of these cross-sectional HRCT images. And then they train by machine learning, to train the software to get a very specific software, which could identify the HRCT picture of the lung and then make the conclusion to predict which the disease is. And on the right-hand side, is the confusion matrix, which shows the comparison between the software’s diagnosis, and the diagnosis done by pharmacists.

The results show that the accuracy of the software is higher than humans. If doctors could have this software, that will be very easy to get the diagnosis, to pick up the patient from a very heavy clinical road, and get a very correct diagnosis.

Sharing and learning

• How does artificial intelligence improve the accuracy of CT image recognition through deep learning technology?
• Symptoms of pulmonary fibrosis may appear after recovery from Covid-19 pneumonia, why do you think this happens?