Artificial intelligence tools may be able to assist doctors in identifying patients who are at risk of developing lung cancer.
To predict the risk of lung cancer, scientists made Sybil, an A.I. deep learning tool. Sybil’s AUC (area under the curve) value was 94%. It shows a high ability to correctly group people with or without lung cancer. This is within a year of testing, and up to 81% within six years. Sybil also reduced the false positivity rate for lung cancer. It went from 14% with current methods of analysis to 8% for the first scan. This allows for a single scan for lung cancer.
Researchers stated that more testing is required to determine Sybil’s performance, particularly in different ethnic groups.
The only chosen method for testing for lung cancer is low-dose computed tomography. They are also known as a low dose CT scan. Patients lie on a table while an X-ray machine creates images of their lungs. Many tested patients do not receive needed long-term care, such as follow-ups. Other research shows that lung cancer diagnoses are rising among nonsmokers and light smokers.
Improving the efficiency of low-dose CT scans and making them available to non smokers and light smokers is an aim. This could reduce lung cancer mortality rates. In addition to 3 or 4 low dose CT scans, current low dose CT scan approaches require a mixture of factors. It needs population data, clinical risk factors, and radiologic annotations for results.
The paper was published in the Journal of Clinical Oncology.
Sybil’s lung cancer prediction:
Sybil only requires one low-chest computed tomography scan to predict lung cancer risk 1-6 years after screening. To encourage further research and clinical applications, Sybil’s algorithm and image annotations are made public.
The team used data from 15,000 participants to create a deep-learning A.I. model. They used 35,001 low-dose CT scans to train and develop their model, and 6,282 to test it.
Two thoracic radiologists analysed suspicious lesions on patient scans. These developed into cancer within a year of the scan to help train the model.
Sybil had a correct rate of lung cancer or not score of 92% across all test data sets after 1 year. 86% after 2 years, and a chance of 75% after 6 years using only single low dose CT scans.
Sybil’s performance was correct across sex, age, and smoking history. Researchers then put Sybil to the test on data from Massachusetts General Hospital (MGH) in Boston and Chang Gung Memorial Hospital (CGMH) in Taiwan. Patients from CGMH did not need a positive smoking history for a low dose CT scan. This was unlike patients from the primary and MGH datasets.
Sybil correctly predicted 86% of lung cancer cases or healthy lungs from the MGH dataset within one year. And 94% of cases from the CGMH data. It also predicted 81% of lung cancers or healthy lungs in the MGH cohort. And 80% in the CGHM cohort after six years.
Sybil could also predict traditional clinical risk factors such as smoking from scans.
Drawbacks of Sybil:
The team mentioned some drawbacks in their model. They noted that 92% of Sybil’s training data came from White patients. This implied that their findings may not apply to more diverse populations.
Additionally, they pointed out that since the training data scans were taken between 2002 and 2004. Advancements in CT technology could have a negative impact on Sybil’s capacity to predict the future.
Conclusions about Sybil’s ability to predict lung cancer in nonsmokers are expiremental. This is because they lacked complete smoking data from CGMH patients. Thus, the researchers concluded that additional assessment in a prospective study is required to gauge Sybil’s effectiveness and clinical benefit.
Parvathy Sukumaran 
