Computer Vision News - February 2023

38 From Journal of Clinical Oncology education. These models can predict risk for up to six years but do not consider CT scans, limiting their performance. Peter and Jeremy knew that a combination of these two ideas could work . By using high-quality 3D volumes, they were sure that signs of cancer, if present, could be extracted . With reasonable accuracy, it is possible to tell someone’s age and smoking history from a CT alone. The development of their deep learning model, Sybil , named after a prophetess in Greek legend and literature, has been four years in the making. “ If you consider the whole pipeline, it feels long, but for Peter and I, the active model development time was much shorter than that, ” Jeremy notes. “ Afterward, it was about getting the right data for validation, using CT this time around introduced new challenges. “ The amount of information you have per patient is more constrained for a mammogram, whereas CTs are big volumes, ” Peter explains. “ Also, mammography is much more standard. A lot of people get mammograms, so you have hundreds of thousands of data points. CT screening isn’t as popular or well adopted. ” The group was inspired by a Google AI model that uses CT scans to predict lung cancer and lung nodule malignancy over a shorter term. Risk models have also been developed to decide who should be screened, considering clinical risk factors, including age and smoking history, and other correlated variables, such as