According to an
article published ahead-of-print in the April issue of the American
Journal of Roentgenology, a Stanford University team has developed a
quantitative framework able to sonographically differentiate between benign and
malignant thyroid nodules at a level comparable to that of expert radiologists,
which may prove useful for establishing a fully automated system of thyroid
nodule triage.
Stanford researcher Alfiia Galimzianova and colleagues retrospectively
collected ultrasound images of 92 biopsy-confirmed nodules, which were
annotated by two expert radiologists using the American College of Radiology’s
Thyroid Imaging Reporting and Data System (TI-RADS).
In the researchers’ framework, nodule features of echogenicity, texture, edge
sharpness, and margin curvature properties were analyzed in a regularized
logistic regression model to predict nodule malignancy. Authenticating their
method with leave-one-out cross-validation, the Stanford team used ROC AUC,
sensitivity, and specificity to compare the framework’s results with those
obtained by six expert annotation-based classifiers.
The AUC of the proposed framework measured 0.828 (95% CI, 0.715–0.942)—”greater
than or comparable,” Galimzianova notes, “to that of the expert
classifiers”—whose AUC values ranged from 0.299 to 0.829 (p =
0.99).
Additionally, in a curative strategy at sensitivity of 1, use of the framework
could have avoided biopsy in 20 of 46 benign nodules—statistically
significantly higher than three expert classifiers. In a conservative strategy at
specificity of 1, the framework could have helped to identify 10 of 46
malignancies—statistically significantly higher than five expert
classifiers.
“Our results confirm the ultimate feasibility of computer-aided diagnostic
systems for thyroid cancer risk estimation,” concludes Galimzianova. “Such
systems could provide second-opinion malignancy risk estimation to clinicians
and ultimately help decrease the number of unnecessary biopsies and surgical
procedures.”