Deep learning improves tumor contouring, may help patients with head and neck cancer
An AI tool improved the primary gross tumor contouring accuracy in patients with nasopharyngeal carcinoma (NPC), according to a March 26 study published in Radiology. The method may help treat patients with the rare form of cancer.
The researchers trained the 3D convolutional neural network on MRI datasets taken from 818 patients with NPC. Two experts manually delineated the gross tumor volumes. The tool was then tested in 203 independent MRI datasets and validated on a cohort of 103 patients.
Overall, the AI method contoured tumors similarly to human experts in 203 patients (median Dice similarity coefficient of 0.79). And in a separate multicenter study comparing the AI platform to eight experienced radiation oncologists, the tool beat 50 percent of experts.
Importantly, NPC can be cured through radiation therapy, wrote first authors Li Lin, of Sun Yat-sen University Cancer Center in South China and Qi Dou, The Chinese University of Hong Kong, and colleagues, but proper contouring of such tumors is central.
“Contouring accuracy is clinically important, as suboptimal tumor coverage and poor-quality radiation therapy plans are major factors for disease relapse and inferior survival,” the authors wrote.
When the radiation oncologists edited contours first generated by the AI tool, five out of the eight experts improved their accuracy, reduced contouring time by nearly 40 percent and reduced intra- and interobserver variation by 36 percent and 55 percent, respectively.
“Our findings show that AI-assistance can effectively improve contouring accuracy and reduce intra- and interobserver variation and contouring time, which could have a positive impact on tumor control and patient survival,” the authors wrote.
The authors noted their application could potentially aid the gross tumor volume recontouring process in adaptive radiation therapy—a known curative therapy for some NPC patients—but they remained cautious of that conclusion.
“While we are enthusiastic to maximize its utility, we must caution the reader that our deep learning algorithm was trained on a multiparametric MRI data set acquired with a 3.0-T scanner, and we are uncertain whether it requires retraining on a CT-based data set,” the authors warned.
In a related editorial written by Zheng Chang, PhD, department of radiation oncology at Duke University Medical Center in Durham, North Carolina, Chang argued rapid advances in AI algorithms will soon make AI-based contouring a valuable tool in radiation therapy. However, more thorough validation must take place before such algorithms are clinically feasible.
“Before the AI contouring tool is fully adopted into clinical use as a part of standard practice, it needs validation in more independent multicenter studies with larger patient cohorts,” Chang wrote. “Although the AI contouring tool shows promising results for NPC primary tumor delineation in this study, section-by-section verification of tumor contour by radiation oncologists should never be omitted.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.