Google Shows AI Can Predict Lung Cancer from CT Scans

Google used advances in 3D volumetric modeling along with datasets from its partners for modeling lung cancer prediction and laying the groundwork for future clinical testing.

Generally, radiologists go through hundreds of 2D images within a single CT scan with cancer being miniscule and hard to spot. Google researchers created a model that can generate the overall lung cancer malignancy prediction (viewed in 3D volume) as well as identify subtle malignant tissue in the lungs (lung nodules). The model can also factor in information from previous scans, which can be useful in predicting lung cancer risk as the growth rate of suspicious lung nodules can be an indicator of malignancy.

The researchers leveraged 45,856 de-identified chest CT screening cases and validated the results with a second dataset and also compared their results against six US board-certified radiologists. They found that when using a single CT scan for diagnosis, their model performed on par or better than the six radiologists and detected 5% more cancer cases while reducing false-positive exams by more than 11% as compared to unassisted radiologists in the study. Google’s approach achieved an AUC (a common metric used in machine learning that provides an aggregate measure for classification performance) of 94.4%.

With only 2-4% of eligible patients in the US being currently screened for lung cancer, Google’s research demonstrates the potential for AI to increase accuracy as well as consistency, which could help accelerate the adoption of lung cancer screening globally. Google now plans to conduct further studies to assess its impact and utility in clinical practice. It is collaborating with Google Cloud Healthcare and Life Sciences team to serve the model through the Cloud Healthcare API and is holding discussions with its partners across the world to continue additional clinical validation research and deployment.

Related Links:
Google