Sybil, the artificial intelligence that detects lung cancer risk

MIT researchers developed an artificial intelligence system to assess the risk of lung cancer in patients, using deep learning.

This tool is capable of analyzing low dose computed tomography images of the lung, known by its acronym LDCT, and predicting the risk of a patient developing lung cancer in the next six years, without the help of a radiologist to perform such an analysis.

Sybil, as the tool is called, achieved C-indices of 0.75, 0.81, and 0.80 over six years from various LDCT images, which means good performance.

Although Sybil is still under development and without a clinical availability date, the researchers are optimistic about the results. The tool would help identify patients at highest risk for the world's deadliest cancer.

The tool was developed with the aim of providing a personalized approach to risk assessment for each patient, which would allow detection at earlier stages, when surgical operations can be performed.

The researchers encountered a major challenge in developing Sybil, because the images used were devoid of any evidence of cancer. Early-stage lung cancer occupies small portions of the entire lung. To do this, they labeled and classified first with images where there were visible cancerous tumors to train Sybil and then with CT images without any signs of cancer.

Once that stage was achieved, the testing phases began with data from the National Lung Cancer Screening Trial (NLST), Mass General Hospital (MGH) and Chang Gung Memorial Hospital (CGMH) and the results were promising, due to the high throughput.

The team has stated that Sybil is in development and not yet ready for clinical use, but they are working on further testing to bring it to market.

Sybil could become a tool to improve early detection of lung cancer and increase the chances of survival.

Read more at the American Society of Clinical Oncology Journal website https://ascopubs.org/doi/full/10.1200/JCO.22.01345