AI Algorithm Detects Lymph Node Metastasis in Cervical Cancer on MRI

July 27, 2020

Algorithm can pre-operatively pinpoint metastasis, potentially helping some patients avoid unnecessary surgery.

A new artificial intelligence (AI) algorithm used with MRI could help patients avoid unnecessary surgery for cervical cancer, a new study reported.

The tool, discussed in an article published July 24 in JAMA Network Open, can identify lymph node metastasis pre-operatively on patients’ scans. It was developed by a team of researchers from the Chinese Academy of Sciences in Beijing, and it melds a deep-learning algorithm with MRI’s ability to assess lymph node status.

“The findings of this study suggest that deep learning can be used as a pre-operative, non-invasive tool for diagnosing lymph node metastasis in cervical cancer,” wrote lead study author Qingxia Wu, Ph.D.

By combining algorithm-gathered tumor details with MRI-identified lymph node status, the hybrid model, with an area under the curve of 0.933, was much better at detecting lymph node metastases. In addition, the researchers determined, this model was also significantly associated with disease-free cervical cancer survival.

Related Content: Evaluating Hybrid PET/MR in Cervical Cancer Patients

Traditionally, MRI is used to measure lymph node size, but it has limited sensitivity for detecting metastasis, frequently making it more difficult to identify the right treatment options for a patient. And, previous attempts to use radiomic features to improve that sensitivity have fallen short – they require time-consuming tumor delineation, and it is frequently difficult to adapt them to specific clinical issues. 

This is where the deep-learning algorithm can step in and be valuable, Wu said. As a non-invasive tool, it can help patients avoid unnecessary surgeries and help providers map out more appropriate treatment programs. It is a tool that does not require precise tumor delineation, and it performs better than radiomic features, he added.

To develop the hybrid model, Wu’s team examined the MRI scans of 338 patients with cervical cancer and used that data to create and validate the models. They, then, tested the algorithm on an independent dataset from 141 patients being treated at Yunnan Cancer Hospital in Kunming, China.

According to their analysis, the team determined a hybrid model that meshed deep learning with an MRI-captured lymph node status produced the best assessment of lymph node metastasis. If the model detected that the largest lymph node seen on MRI had a diameter of greater than or equal to 1 cm diameter, the node was given a positive MRI status. That result was, then, used to provide an “H score” – a predicted lymph node metastasis probability score.

Alongside predicted metastasis, Wu explained, the model was also effective in pinpointing a patient’s cervical cancer prognosis. Not only were patients with higher H scores considered high risk, but they also had a lower probability of being disease-free. And, then when they were, they had shorter disease-free survival times than patients with lower H scores.

Although the study did have limitations and required further research to enable generalized performance, the team concluded their model could play a beneficial role in the detection of lymph node metastasis with cervical cancer patients.

“Our study develops an end-to-end [deep-learning] model to detect [lung nodule metastasis] during routine MRI,” the team wrote, reiterating the problems with existing assessment methods. “We developed a [deep-learning] model to try to overcome these problems by automatically learning [lung nodule metastasis]-related features, providing a helpful adjunct to assess [lung nodule metastasis.]