Computer programs could help provide more accurate prognoses for people with cancer.
For women with cervical cancer, a recurrence of the primary tumor can be as worrisome as metastasis, for the chance of a cure is often minimal. “One woman’s cervical carcinoma will respond to therapy but another’s won’t, and we don’t know why,” observes Nina A. Mayr, professor of radiation oncology at Ohio State University’s James Cancer Hospital and Solove Research Institute. “But when a patient looks you in the eye and asks what are the chances that treatment will indeed work for her cancer, telling her ‘I don’t know’ isn’t very satisfactory.”
Mayr has hope that a computer model being developed may someday predict whether her patients’ cervical cancer will return. Her colleagues Jian Z. Wang and Zhibin Huang studied 80 women undergoing radiation therapy for cervical cancer, using MRIs to measure the volume of the tumors as they shrank. After following the women for six years, they calculated that two factors were primarily responsible for the women’s varied recurrence rates: the portion of cancerous cells that survived daily radiation treatments and the time it took each woman’s body to clear out the destroyed cells. By plugging the data into a computer model, they were able to estimate a tumor’s degree of resistance to radiation therapy, and thus the woman’s chances of recurrence.
Given advance warning of a tumor’s slow response to treatment, physicians may be more willing to try radical treatments early on. “Right now, we wait until conventional therapy fails before doing something as drastic as using higher doses of radiation, and by then it may be too late,” says Mayr. “With this knowledge, we can individualize treatment and we might improve survival.”
Researcher Karen Drukker of the University of Chicago has also designed a computer program to help radiologists; hers, however, uses artificial intelligence to help predict the spread of breast cancer. The first stop for migrating breast cancer cells is the lymph nodes, where they establish micrometastases. But in the early stages of cancer spread, the diagnostic ultrasound that radiologists use to probe lymph nodes after a breast tumor is found often reveals no sign of metastasis.
Drukker found that three characteristics of the ultrasound images of a woman’s primary breast tumor helped predict which lesions will metastasize. In a pilot study that reanalyzed the ultrasounds of 50 women who had suspected breast cancer and who had seemingly normal lymph nodes on the ultrasound, the computer program correctly identified most of the 20 women whose lymph nodes later revealed metastases upon biopsy or surgery. The next step is testing whether several radiologists using the program will be able to more accurately diagnose metastases.