From the Journals

AI improves diagnostic accuracy in cervical cancer


 

A deep learning (DL) computer model improved upon the accuracy of cervical cancer diagnoses compared to traditional radiology. This could allow some women to avoid surgery and be treated with chemotherapy instead, suggested researchers.

The model mined tumor information from pelvic sagittal contrast-enhanced T1-weighted MRIs and combined this with clinical MRI lymph node status.

It was 90.62% sensitive and 87.16% specific for predicting lymph node metastases (LNMs) in a validation cohort of women who underwent surgery for cervical cancer.

The area under the curve was 0.933. The approach was significantly associated with disease-free survival (hazard ratio, 4.59; 95% confidence interval, 2.04-10.31; P < .001).

The study was published online on July 24 in JAMA Network Open.

“The findings of this study suggest that deep learning can be used as a preoperative noninvasive tool to diagnose lymph node metastasis in cervical cancer ... This model might be used preoperatively to help gynecologists make decisions,” said investigators led by Qingxia Wu, PhD, of the Northeastern University College of Medicine and Biomedical Information Engineering in Shenyang, China.

“Studies like these suggest that deep learning has the potential to improve the way we care for our patients,” but there’s much to be done “before these types of algorithms will be commonplace,” commented Christiaan Rees, MD, PhD, an internal medicine resident at Brigham and Women’s Hospital, Boston, who has a doctorate in quantitative biomedical sciences.

Next steps include repeated validation across multiple control groups, he said in an interview, as well as “finding ways to effectively integrate these tools into the radiologist’s day-to-day practice. One possibility would be for direct integration of the algorithm into the electronic health record.”

Accurate prediction could lead to skipping surgery

Chemotherapy, rather than surgery, is an option for women with positive lymph nodes (LNs), so accurate prediction can help them avoid an operation and its risks, the authors said.

The problem is that “the traditional methods for assessing LN status in cervical cancer, which rely mainly on assessing the size of LNs on MRI, have limited sensitivity in diagnosing LNM in cervical cancer and might lead to inappropriate treatment decisions,” they wrote.

“Although sentinel LN dissection ... shows good sensitivity and specificity, its application is limited by available facilities and experts,” the team said.

DL is an advanced form of artificial intelligence in which a computer program continuously improves on a given task as it incorporates more data – in Dr. Wu’s study, more than 14 million images. Deep learning has recently shown promise in several imaging tasks, such as diagnosing Alzheimer’s disease and screening for breast cancer.

Once adapted for cervical cancer, DL “does not require precise tumor delineation, making it an easy-to-use method in clinical practice. In many tumor analysis tasks, DL outperforms traditional radiomic features,” the team noted.

The study involved 479 women – 338 during model development, and 141 in the validation cohorts. The mean age of the participants was 49.1 years. They had undergone radical hysterectomy with pelvic lymphadenectomy for stage IB-IIB cervical cancer within 2 weeks of a pelvic MRI. Pathology reports were used to check the accuracy of the model’s predictions.

Specificity, sensitivity, and area under the curve were a little better in the study’s development cohort than its validation group, for whom median disease-free survival was 23 months versus 31 months among the patients in the development cohort. Nodes were positive on lymphadenectomy in a little more than 20% of women in both groups.

Incorporation of both intratumoral and peritumoral regions on contrast-enhanced T1-weighted MRIs versus axial T2-weighted and axial diffusion-weighted imaging, produced the highest sensitivity. Adding MRI-LN status – defined as positive when the short-axis diameter of the largest LN on MRI was ≥1 cm – improved the model’s specificity.

To understand how the model reached its conclusions, the team analyzed how it extracted features from tumor images. “In the shallow convolution layers, the DL model extracted simple tumor edge features ... while in deeper convolution layers, it extracted complex tumor texture information ... In the last convolution layer, the DL model extracted high-level abstract features (the fourteenth layer). Although these high-level features were so intricate that they were hard to interpret by general gross observation, they were associated with LN status,” the investigators said.

The team notes that “both intratumoral and peritumoral regions were necessary for the DL model to make decisions,” which “can probably be explained by the fact that higher lymphatic vessel density in peritumoral regions might lead to higher regional LNM.”

Commenting on the study, Dr. Rees said that “the authors did a [good] job of essentially deconstructing their neural network to see what the algorithm was actually picking up on to make its decision.

“One of the nice features of deep learning is that once the algorithm has been developed and validated, the end user doesn’t need any experience in deep learning in order to use it,” he added.

Even so, “while these resources can be incredibly powerful tools, they should not function in a vacuum without human judgment,” Dr. Rees said.

The work was funded by the National Natural Science Foundation of China, among others. The investigators have disclosed no relevant financial relationships.

This article first appeared on Medscape.com.

Next Article: