Deep learning can be deployed as a preoperative tool to assess if cervical cancer has spread to the lymph nodes, potentially sparing patients from unnecessary surgery.
A team of researchers from China and Texas made this discovery after creating models from hundreds of patients with the disease who underwent an MRI prior to surgery. A hybrid approach combining tumor information and imaging data identified more than 90% of metastatic lymph node cases, the group reported Friday in JAMA Network Open.
Given that cervical cancer is one of the most common cancers among women, these results may help personalize care plans and avoid invasive options for many struggling with treatment decisions, Qingxia Wu, PhD, with Northeastern University’s College of Medicine in Shenyang, China, and colleagues wrote.
“Specifically, patients who show evidence of LNM may undergo chemoradiotherapy rather than surgery as their first choice, avoiding surgery followed by adjuvant chemoradiotherapy and possible serious complications thenceforth,” the group added. “Therefore, accurate identification of LN status preoperatively in patients with cervical cancer might avoid unnecessary surgical intervention and benefit treatment planning.”
Wu et al. built their deep learning platforms from a primary group of 338 patients from two Chinese hospitals and evaluated the models on an independent cohort of 141 from a third location.
The approach that harnessed both information from within and directly surrounding the tumor on contrast-enhanced T1-weighted imaging proved to be the best at identifying lymph node metastasis (area under the curve of 0.84).
A hybrid approach, combining tumor image data and MRI-reported lymph node status, however, improved these results (AUC of 0.93). The latter also notched a sensitivity of 90.6% and specificity of 87.2%.
“With the apparent high sensitivity and specificity of our hybrid model, this model might be used preoperatively to help gynecologists make decisions,” the authors wrote on July 24.
Wu and colleagues cited few limitations of their research but noted more extensive and prospective data will be needed before their model can be applied to more patients.