AI may soon be used to screen for pancreas cancer, Elizabeth Tracey reports


Because it is most often detected late, pancreas cancer has a very poor prognosis. Now a new study used common CT images without a material known as contrast to enhance them, trained a computer on them and showed impressive detection. William Nelson, director of the Kimmel Cancer Center at Johns Hopkins, explains.

Nelson:    The reason they looked at noncontrast images were those are the kinds of images that you may have gotten for another purpose and would certainly be more likely to look like this screening images that you might imagine someone taking. They trained it to be able to recognize seven different types of tumors. How well did it work? They looked at another dataset of 6200 and another dataset that was substantially larger. They compared this to radiologists with the caveat that they’re not trained to look for these tumors on noncontrast scans, they had a sensitivity of almost 93% and a high level of specificity, which outperformed the human radiologists.  :35

Nelson says hopes are high that such a practical, AI driven approach will help reduce the toll of this deadly disease. At Johns Hopkins, I’m Elizabeth Tracey.