It’s safe to say that artificial intelligence (AI) is at the forefront of medicine. Research teams are figuring out how to harness AI’s power to do everything from discovering strokes, predicting someone’s 5-year risk of breast cancer, distinguish among pancreatic diseases, and virtually everything in between.
Another avenue that AI experts are exploring is thyroid cancer screening. In the United States, thyroid cancer’s incidence has tripled in the last 30 years, though experts believe this is probably because specialists are using ultrasounds to screen their patients more often; ultrasounds can pick up small, potentially cancerous nodules that weren’t previously detectable.
In late October, a team at Philadelphia’s Thomas Jefferson University led by Elizabeth Cottrill published a study in which they looked into whether pairing AI with the thyroid ultrasound could provide patients with a rapid, non-invasive way to screen for thyroid cancer. The current gold standard is that if a thyroid nodule is detected, a doctor will recommend a biopsy as a follow-up to further assess the situation. The results, however, are not always conclusive.
In a statement, Cottrill explained that “Currently, ultrasounds can tell us if a nodule looks suspicious, and then the decision is made whether to do a needle biopsy or not. But fine-needle biopsies only act as a peephole–they don’t tell us the whole picture. As a result, some biopsies return inconclusive results for whether or not the nodule may be malignant.”
The discoveries made from Cottrill’s study suggest that pairing AI with ultrasound could provide specialists with a more effective way to separate high and low risk nodules in the thyroid. As many as 67% of the general population has thyroid nodules, and though most are harmless, an ultrasound must be performed on one if it is detected. Physicians often recommend biopsies be performed on nodules that are larger than 1 cm, and some are even evaluated using molecular genetic testing, which is expensive but less invasive than a biopsy.
Senior co-author John Eisenbrey stated that AI provides an efficient, low-cost way to assist physicians in analyzing an “indeterminate” nodule, or one that is difficult to come to a conclusion regarding.
The research team developed an AI using a Google platform that could be used alongside a thyroid ultrasound with one primary goal: to be used in conjunction as a cheap and rapid primary screen for thyroid cancer. To train the algorithm, they used 556 images of thyroid nodules collected from 121 patients who underwent fine-needle biopsies guided by ultrasound followed by molecular testing. The molecular testing categorized 91 of the nodules as low risk, and 43 as high risk; the results were based on a specific gene panel.
Afterwards, the research team tested their AI on a separate set comprised of 53 images to see how good it would be at classifying thyroid nodules in comparison to results from molecular tests.
The results were great: the AI algorithm had an overall accuracy of 77.4% and a specificity of 97%; the latter result indicates that there would be a low rate of false positives. While the results of this study are considered preliminary, they suggest that this AI could be a useful tool in helping diagnose thyroid cancer more accurately.
