MIT develops AI that predicts lung cancer years in advance
3 Min Read
The newly developed tool accurately predicts the risk of lung cancer in people who have or have not smoked in the past.
-
AI tool to predict lung cancer. (Getty Images)
A new AI deep learning model has been newly developed by researchers at MIT (Massachusetts Institute of Technology) to predict lung cancer risks up to six years in advance through a single low-dose CT scan.
The modelÙˆ known as "Sybil"Ùˆ is intended to use a single low-dose chest scan to predict the risk of lung cancers occurring 1-6 years after a screening, in contrast to current lung cancer prediction models that require a combination of demographic data, clinical risk factors, and radiologic annotations.
The overall process of lung cancer screening has been compared by co-first author and PhD student at MIT Peter Mikhael to "trying to find a needle in a haystack."
Read next: US FDA approves AstraZeneca's lung cancer drugs
However, the study revealed Sybil was able to predict both short-term and long-term lung cancer risk, earning C-indices scores ranging from 0.75 to 0.80Ùˆ by utilizing a diverse set of scans from two hospitals and the National Lung Cancer Screening Trial. Strong models are indicated by values over 0.8.
The model was even more successful at predicting cancer risk one year in advance, with values on a ROC-AUC probability curve ranging from 0.86 to 0.94.
No visible cancer on the scans
Due to the fact that early-stage lung cancer only takes up a small portion of the lung or a small portion of the hundreds of thousands of pixels that make up each CT scan, the imaging data used to train Sybil was largely free of any indications of cancer.
Lung nodules, which are denser regions of the lung tissue, have the potential to be cancerous, but the majority are not and instead are healed infections or airborne irritants.
Read next: New cancer treatment offers hope to patients out of options
Co-author Jeremy Wohlwend was taken aback by Sybil's impressive performance despite the absence of any palpable cancer. “We found that while we as humans couldn’t quite see where the cancer was, the model could still have some predictive power as to which lung would eventually develop cancer.”
The study was conducted at the Jameel Clinic at MIT under the direction of Professor Regina BarzilayÙˆ and was published in the Journal of Clinical Oncology in collaboration with Chang Gung Memorial Hospital in Taiwan and Mass General Cancer Center.
